The fifteen desktop modules' tools tests failed on every manifest with
"json: unknown field class": a test helper copied into each module decoded
module.json into its own struct, refusing unknown fields. Manifest
strictness is the catalog check's alone, and the manifest's form is a
contract read through mesh-sdk's catalogue package, so the copied
`manifest` struct and its reader go from the fifteen desktop modules and
pacman; each test reads module.json with catalogue.Decode and asserts only
the module's own promises (claims, files, tools) on catalogue.Manifest.
Builds: design 47 H2 H-A-498, H-A-498 addendum
The public proxy serves the forge's paths from git tags and direct reaches
the forge's git, so a version missing from the registry was found elsewhere
instead of failing loudly. The fetch now downloads in two passes over every
directory: first each pinned mesh package with GOPROXY=<seat>,off (a miss says
"the Go package registry holds no <pkg>@<v>"; no binding refuses "no Go
package registry bound"), and every zero-version requirement of the graph is
taken or refused then; second the rest of the build list with the public
proxy alone, no direct. Before either, a go.mod naming a mesh package at a
version other than the zero version is refused, so no forge path can reach
the public proxy; a pull request head is placed only for an ask's commit, and
the placement of go.mod-named pins goes. Every other go command (test, lint,
compile) runs with GOPROXY=off; GOPRIVATE and the account's ssh settings leave
goEnv, as repository fetches set their own.
Builds: design 47 H2 H-A-494, H-A-494 addendum (two downloads), H-A-494 second addendum (a forge path at a real version is refused), H-A-494 third addendum (a pull request head places only an ask's commit pin); design 51 Go packages (the recipe test refuses direct)
The operator typed `nox licence` and got the retired controller's refusal
instead of the licences. `nox licence` and `nox licence list` now call the
licence manager's licences and bindings and print one table of each, every
time local (1126/AC1); `nox licence usage [<licence>]` prints the usage
readings as a table (1126/AC2). `--json` prints the raw answer on every
rendered line, `nox nodes` included: one behaviour for one flag; for the
two calls one object {licences, bindings} (1126/AC3).
In the same change, the operator (2026-11-22 20:41): "why are there still
references to the controller?". Every retired controller line nox answered
with "this was the controller's" (its `instead` lines and the `unowned`
map) is deleted with its text: none has a seat verb of the same job; a
retired line now answers "is not a command" with exit 2, and nothing nox
prints names the controller (1126/AC4). The comments, help and README no
longer name the controller or issue 1126.
Builds: issue 1126 note 151, answers 1-5, and the note after it (acceptance
1126/AC1..AC4, the operator's 20:41 line); one-question-one-tool;
principle 4, a module is steered by its tools.
A mesh package's tarball is unpacked straight into node_modules, so its own
dependencies were never there: the forge module's bundle failed with esbuild
"Could not resolve blakejs" from the SDK's seal code. After every mesh package
is placed, the fetch installs the dependencies they declare (not their
devDependencies, not a mesh package, not one already present) through the
same npm path as the module's own, before the bundle.
Builds: design 47 H H-A-491 (addendum)
The Node build module judged its levels by the same fixed bounds, and its
integration setup took 40 s against 30 s while three nodes' checks ran.
The rule is the kind's, not the Go module's: its bounds, and the
MESH_TEST_LEVEL_BOUND its scripts get, are now their defaults times the
steps the seat has in hand at the step's start (at least 1), and the
result names them (bound, setup_bound).
Builds: design 47 H2 H-A-488, H-A-488 addendum
On a loaded build node the integration level's setup took 34-37 s against
its 30 s bound three times today: the node-engine twice, the assigner once.
The bounds were judging the node's load, not the tests. A level's bounds,
and the MESH_TEST_LEVEL_BOUND its tests get, are now their defaults times
the steps the seat has in hand at the step's start (at least 1). The step's
result keeps naming the bounds it was judged by (bound, setup_bound).
Builds: design 47 H2 H-A-488
The Node build module kept only the last half of its bound of a failed
command's stderr, so esbuild's error line was pushed out by its Node
launcher's stack trace ("esbuild for code: status 1" with no cause). The
image build module's reason and journal line kept stderr's tail alone too,
and so did the Node module's toolchain install. Each now keeps the head and
the tail within its bound, with "… <n> bytes left out …" between them. The
Go build module keeps a failure's first line, not its tail, and is left as
it is.
Builds: design 47 H2 H-A-475
Held compared the version tag's manifest digest with a bundle's or an
archive's blob digest, so every second publish of such a version was
refused even at the same bytes. It now compares like with like: a bundle's
or an archive's stored layer, an image's manifest. The same digest is
already held (ended, published said, nothing stored), on a repeated ask too;
another digest is refused naming the held one. The fake store keeps a
manifest digest beside the layer's, and a row runs the module's OCI code
over a store in memory.
Builds: design 47 H2 H-A-474, design 47 H H-A-393
Review round 1: a durable whose twin is current was removed without
reading the durable itself, so of the five live pairs where both twins
have pulls waiting either one would have gone. The tool now reads the
named durable too and refuses it while pulls wait on it, both read from
the bus at the call. "Current" is the setting twin-current-hours (24 by
default) of the manifest, read from the configuration at each call, the
field name the grants part's manifest names too.
Builds: design 47 H2 H-A-472 (1) and its third addendum (1), (2); review round 1 findings (1), (2)
An unread or stale view held no entry for a watched repository, so every
member would have merged plainly with the note no delivery: a merge that
reports success and delivers nothing. The registrations holder now keeps
the time of its last read and of the last registration-changed heard; the
merge call refuses every member ("the watched branches are not read yet")
while the view is not read since that change, and a read that fails is
said and asked again with the module's growing wait until it answers.
Builds: design 47 H2 H-A-471 (second addendum); review finding H-A-471 addendum (2)
H2 set finding-events to max-mib 8 on every module that binds it; the
healers' manifest test still expected 16 MiB and failed at main.
Builds: design 47 H2 §783, §812; H-A-411 (1), H-A-430 (3)
The operator wants the merge call to work for a repository of no module,
such as hq. A member whose repository has no watched branch (no entry, or
an empty set, in the registrations' watched branches) is judged against
nothing of the module graph (no failed required step at a fingerprint),
merged in the order named, answered and kept merged with the note "no
delivery: the repository has no watched branch", and left out of
merge-called by that kept note, at the call and at the start's re-say, so
it starts no delivery. A group mixing it with watched members judges and
delivers those as before. A watched repository whose base is unwatched
stays refused. The holder answers at the seat git's form 3.
Builds: design 47 H2 H-A-471 and its addendum (1), (2)
A module that moved into the tool runner left its program's durable
<node>_<module>, and one that runs its own program again left
<node>_<module>_tools: about 110 such orphans the tool refused by name.
It now removes a module's durable when its twin on the same stream has
pulls waiting and a delivery within the day, and names the twin in the
hand act; a module's only durable stays refused.
Builds: design 47 H2 H-A-472 (1), H-A-127, H-A-128, G A-340
The entries whose refusal a record's end must clear were kept in process
memory, so after a restart a refusal of an entry never written (an act's
first write, a mover's) was never cleared. A record-too-large finding now
names its record as its subject (its id stays the refused entry), and the
record's end clears every held one whose subject is the record, read from
the gate's own findings state; the in-memory map goes. A finding raised
before this change names its entry as its subject and clears when that
entry is next written within 64 KiB.
Review finding: record-too-large cleared at the record's end only from
process memory (movers.go:228-233, Records.refused).
Builds: design 47 H2 §809, H-A-455 (3), H-A-460
An act the gate owes the forge is now the entry act/<branch.source.act.id>
(type act, index acts on its record's key in the manifest), removed when
the forge bridge performed it; the head entry holds the record's fields
alone, and the record's acts are joined from their entries when read.
The record type keeps no act operation: a head written before this change
is read for its fields and the acts embedded in it are ignored (they are
notes and views of the few records in flight at the delivery; losing them
changes nothing the operator relies on). A view act's text is the record's
summary, the head's fields and a count of movers per outcome, so it fits
64 KiB for any record; gate_record still shows the whole record. The
64 KiB refusal measures every write, the head's, a mover's and an act's,
raising record-too-large, which the record's end clears for every entry
of it. The gate's whole test gains the step: 126 movers write 127 entries
plus their acts, none over 64 KiB, the head counting 126 and holding no
act; a take-in said again with a why past 64 KiB raises record-too-large
and the record's end clears it.
Builds: design 47 H2 §809, H-A-385, H-A-391, H-A-453 (3), H-A-455, H-A-457, H-A-459
The seat git's whole test configures no token: the bundle mints the tools'
account's token at the fixture forge with the admin's login, writes it
through the vault seat's two-step own-put (sealed to the call's key) and
reads it through get (sealed to the process's key), both answered by the
test, with the mint's lease watched as the runtime delivers it. After the
mint the forge holds exactly one token for the account, mesh-tools-1, the
one the vault keeps at generation 1; the old fixed-name token is gone.
Builds: design 47 H2 §808 (whole: the forge holds one token for the account after), H-A-422, H-A-430 (1), H-A-439, H-A-453 (3)
The review found §810's probe proven only at the unit level (H-A-453 (3)): the whole test now enables the budget
probe, answers gate_times with a core delivery 11 minutes from its merge call, reads over-budget for core naming its
record and longest phase, and a core delivery within its budget at the next run clears it.
Builds: design 47 H2 §810, H-A-453 (3)
test/fakebus.ts imports tweetnacl to seal as the vault does; it came only through mesh-sdk's own dependency, so the
forge module's tests named nothing they import (the review's finding; H-A-453). Pinned as mesh-sdk pins it, an
external package as the rule allows.
Builds: design 47 H2 §808, §822; H-A-407, H-A-453
Read on novox at 2026-11-22 01:35 UTC through the bus module's nats_streams
(the review's finding on H-A-437; H-A-453):
gitea bound 16703 bytes, 28 msgs cap 8 MiB
gitea announced 98 bytes, 1 msg cap 8 MiB
gitea findings 565816 bytes, 2013 msgs cap 8 MiB
gitea first-views 369 bytes, 1 msg cap 8 MiB
gitea transfers 396894 bytes, 2015 msgs cap 8 MiB
gitea idle 376879 bytes, 1041 msgs cap 8 MiB
gitea failed-steps 255007 bytes, 754 msgs cap 8 MiB
gitea calls 36362 bytes, 93 msgs cap 8 MiB
gitea taken no bucket on the bus yet cap 8 MiB
gitea-findings log 0 bytes, 0 msgs cap 8 MiB
Each fill is below a quarter of 8 MiB (2 MiB), the largest 565,816 bytes, so
each cap stays at 8 MiB; none is raised.
Builds: design 47 H2 §812; H-A-411 (2), H-A-437, H-A-453
The review found §810's whole criterion proven only at the unit level (H-A-453 (3)): the whole test now ends its
record through the wants seat's wanted and pipeline-wanted and the node-engine seat's running, over the real bus, and
reads the head's delivery-time and class core from a core mover's catalog entry.
Builds: design 47 H2 §810, H-A-453 (3)
The forge module's internal/seatgit whole test needs the Node build seat, and no Go artifact's from holds it, so
nothing named its need and the Go build seat ran it on a node without the Node toolchain. The manifest now says the
need at its top level, read by the Go build module for such packages; gitea declares node-build-node
(review round 1, mesh-catalog, finding on §800; H-A-453 (2)).
Builds: design 47 H2 §800, H-A-453 (2)
A pull request closed unmerged is a fact of the forge, not a fault of the
act, so the gate no longer keeps a refusal for an act the forge bridge
ends not-needed: the record's acts and gate_record show it ended with
that outcome, and delivery_status says no refusal of it. Every other
outcome other than performed is still said as a refusal.
Builds: design 47 H2 §807; H-A-432 (2), H-A-444
The side branch's manifest rows, gate, forge, packet-filter and
package-manager changes join the build modules' share; no file was
touched by both.
Builds: design 47 H2 (the merged commits' own Builds: lines)
The forge bridge ends an act whose pull request closed unmerged with the
outcome not-needed. The gate keeps it in the record's acts as ended, as it
keeps performed and could not be performed, and no longer counts it as
owed. The gate's code takes any outcome forgemsg reads, so the change is
forgemsg's outcome list at mesh-delivery feat/47-H2-a. The row in the
owed-acts test fails where forgemsg refuses not-needed.
Builds: design 47 H2 §807; H-A-432 (2)
The file beside the token that kept its scopes went with C4, and with it
the re-mint of a token carrying a scope beyond the tools' (write:admin).
The value written through own-put now holds the token and its scopes; a
token minted with other scopes, or a value naming none, is minted again
unasked and the one it replaces deleted at the forge. The Go bundles read
the token out of that value. The lost test is back as a row of the scope
test in test/token.test.ts.
Builds: design 47 H2 §808; H-A-439
A bundle publishes under the version its ask reserved, and the delivery SDK's
publish refuses an empty one loudly; the tests' bundle asks carry a version, a
bundle asked with none is shown refused, and an archive still publishes by its
digest alone.
Builds: design 47 H2 H-A-373, H-A-373 addendum
The gate's owed event carries record, the delivery record's source
revision <branch>.<source> (the record's key), said and kept in the
record's acts alike, so the forge bridge keeps it with the act and
bridge_owed_again finds a record's acts in its own state. Built against
the seat mesh-delivery's Owed.record on mesh-delivery feat/47-H2-a.
Builds: design 47 H2 §807; H-A-432 (1)
Read on novox at 2026-11-22 (the seat mesh-broker's show; the gate's
old-section state source through the bus module's nats_streams, since the
bus module made no entry for it):
mesh-delivery-gate heads 582 bytes, 5 keys cap 8 MiB
mesh-delivery-gate findings 24905 bytes, 27 keys cap 8 MiB
mesh-delivery-gate source 70166 bytes, 68 msgs cap 8 MiB
artifact-store-tools findings 8199 bytes, 8 keys cap 8 MiB
Each fill is below a quarter of 8 MiB (2 MiB), so each cap M1 wrote stays
at 8 MiB; none is raised.
Builds: design 47 H2 §812; H-A-411 (2), H-A-437
The packet filter holder's tools read allowances-max themselves (config at
tools/index.ts); no file places it, so the manifest says "runtime": true
and the judge's placed-nowhere rule holds it as read by the module's code.
Builds: design 47 H2 H-A-430 (4)
The eleven other modules that bind finding-events (claude-code, mesh-healers,
mesh-knowledge, mesh-provisioner, mesh-recovery, mesh-self-check,
mesh-watcher, minio, nfs-server, restic, zfs) set max-mib 8 for 7 days, as
the delivery modules' already do, so the indexer's finding-events rule (I6)
judges every manifest alike.
Builds: design 47 H2 §783; H-A-411 (1), H-A-430 (3)
Two processes that minted at once left the forge two tokens by one name, and a token kept in a file in the
runtime state is a plain state. Both accounts' tokens (mesh-tools, mesh-packages) are now the module's own
secrets in the vault, written through the seat mesh-vault's own-put and read through its get, sealed to the
reading process's key. A refused or expired (token-ttl) token is minted again only by the holder of the lease
mint in the state token-lease (mint-lease); each mint is named <account>-<generation>, and the token it
replaces is deleted by id only after own-put landed. The Go bundles read their tokens through get at start
and at secret-changed; the MESH_GITEA_*_TOKEN_FILE words go. forge_token_lease reads the lease and each
token's age, never the token.
Builds: design 47 H2 §808, §822; H-A-407, H-A-422, H-A-426, H-A-430 (1)
Each of the four is core. Every state that named no cap names one, from its
live fill read 2026-11-22 01:4x CET (mesh-broker show, nats_streams on novox):
mesh-build-go steps 16.0 MiB (16,783,695 bytes) -> 128 MiB; mesh-build-node
steps 2.35 MiB (2,463,833 bytes) -> 16 MiB; every other state under 2 MiB ->
8 MiB (largest: distribution holds 1,844,287 bytes, mesh-build-go
detected-steps 868,554, distribution transfers 846,359, findings 668,161, idle
401,736, mesh-build-go pauses 254,366). The Go build module's finding-events is
8 MiB for 7 days. The objects stores already name their caps; none keeps a log.
The Go and Node build modules prune their state detected-steps (each key an
artifact's steps as detected, detected again when gone) at their own writes,
past prune-mark 70 % down to prune-target 50 %, and serve
<module>_state_prune_now and <module>_state_keep; no other state of the four is
prunable (each is truth: running steps, pauses, keys, toolchains, findings,
holds, transfers, windows, mirrors, taken asks).
Builds: design 47 H2 §812, §820; H-A-411 (1), (2), (3), H-A-437
The Go build module's level answers its text in output (test2json's Output
lines), the Node build module's what its commands wrote; the delivery SDK's step
run keeps it as <node>.<ask> in the module's outputs objects, which each module
asks for in requires (1 GiB, kept 7 days) and reads through the verb output at
form 2. Each module's kit test runs the SDK's SeatVerbs.
Builds: design 47 H2 §791, §820; H-A-412 (1), H-A-412 second addendum (a), H-A-421, H-A-434
The objects template takes max-age-days (absent keeps an object for good): the
keeper plans it as the object store's max age, which the bus applies as the
store's TTL, so a build module's outputs objects go at their seven days.
Builds: design 47 H2 §820 (outputs: 7 days, the store's max-age), §791; H-A-434
An install ask's body runs npm ci from the lock with no install script and no
link, packs the tree alone under modules/<module>/node_modules, publishes it to
the artifact store as the archive <module>:node-modules under the ask's version
and answers its reference, which the step's end carries. The package.json npm
ci needs is written from the lock's root entry, a lock without one refused. The
integration level's setup bound in code goes: half of the level's bound, which
each level's script gets as MESH_TEST_LEVEL_BOUND.
Builds: design 47 H2 §800; H-A-399 (2), H-A-403, H-A-427
The delivery SDK replaced the output chunks with the verb answer over the
steps state's objects and output at form 2: each build module keeps an end
over the steps state's value with KeepAnswerIn, serves AnswerTool and
OutputTool over its steps state, and its steps name their outputs objects.
Builds: design 47 H2 §791, H-A-412 addendum, H-A-412 second addendum (b), H-A-399 (1)
The check counts every test the integration level runs, untagged ones in: the NFS server
module held 51 and the package manager module 121, over the budget of 50. Each folded file's
tests run in order as rows of one test that keeps the first one's name and logs each case
before it runs; the root helper's test sockets are numbered per call, as one test now holds
several listeners, and the case that skips without pacman-conf runs last. Counts now 48 and 42.
Builds: design 47 H2 §819; H-A-349 DQ-14
Each build module's steps read tool listed only `<node>.<id>` keys and so
skipped every ask's step, kept as `<asker>.<node>.<id>`: it now reads every key
of its node, newest id first. The Go build module's test of capacity 0 also
holds the seat's current at form 2: no ask taken, why not capacity 0, then
paused.
Builds: design 47 H2 §801; H-A-399 (1)
The Go build module's level result carries not-run (the SDK reader's NOT RUN
skips), its tests get MESH_TEST_LEVEL_BOUND, the level's bound, from which
mesh-sdk's helpers take the setup bound; a test call naming an install archive
reads it from the artifact store by its reference and unpacks it into the
workspace's tree before go test. check-needs is read from the module's
manifest, on the Go artifact whose path holds the package, and the marker
file goes: the Node build module's manifest declares it on its bundle.
Builds: design 47 H2 §800; H-A-294, H-A-350 DMC-6, H-A-368, H-A-399 (1) (2) (3)
The manifest validator takes no # in consumes, and a key part escapes its
dots (mesh-sdk keys.EscapeName), so the forge module keeps hearing
step-ended.*.*.*: a source-only check's module source.<owner>.<name> is one
escaped subject part, and its body names it whole. The whole test says it so.
Builds: design 47 H2 §815; H-A-408
The runner's step-ended of a source-only check names the registration's key
source.<owner>.<name> as its module and the commit's root tree hash as its
fingerprint, so its subject holds more parts than a module's: the forge
module hears every subject under step-ended and judges it as a module's.
The seat's whole test refuses an hq pull request whose unit step failed.
Builds: design 47 H2 §815; H-A-408
heads is written through the SDK's pruned state, which judges the fill each
put answers and past prune-mark (70 %) prunes the oldest keys to
prune-target (50 %); the gate serves mesh-delivery-gate_state_prune_now and
mesh-delivery-gate_state_keep, each written in the hand-act log.
Builds: design 47 H2 §812; H-A-380, H-A-386, H-A-411 (3)
A module that claims a seat is core, one that claims none a leaf (141
manifests; the build modules and the registry are the other developer's).
The forge module's, the gate's and the sweep's states each name 8 MiB, and
their findings stream 8 MiB for 7 days.
Builds: design 47 H2 §810, §812, §783, §820; H-A-394 (1), H-A-411 (1), (2), H-A-378
The forge module reads ask-taken-bound from MESH_CONFIG_ASK_TAKEN_BOUND, the
word the node-engine composes for every setting a manifest declares; the
check takes MESH_CONFIG_<KEY> of each declared config key as set, and an
undeclared one as set by nothing.
Builds: design 47 H2 §775, §781; H-A-370, H-A-400 (2)
A derived accept rule's one comment says derived <modules> <protocol>/<port>
scope <any|private-network|local-network>, its module list cut with a count
past 128 bytes; allow without a source is refused for a port whose derived
rules name no scope any, read from the comments, never from a rule's shape.
Builds: design 47 H2 §813 (4); H-A-357, H-A-384 (1), H-A-349 DQ-11
From..to inclusive, each read from the forge one by one, with its merge
commit, when it was merged and the merge call that made it when a kept call
names it merged; at most 768 KB, capped with the next number as the cursor.
Builds: design 47 H2 §812, §811; H-A-402
Each act (status, comment, view, note, merge-pull, tag) carrying an ask id is
taken in the state taken, key git.<ask>, for the setting ask-taken-bound
(10m, read from MESH_CONFIG_ASK_TAKEN_BOUND): a repeat while the act runs
joins it; after it ended, the repeat answers what the forge holds of the act,
read afresh (a merge call's pull requests merged with their merge commits),
and the act is never done twice. An act that performed nothing drops its
key, so the kept call's repeat acts.
Builds: design 47 H2 §775, §820; H-A-381, H-A-400, H-A-359a
oci-artifact-registry's list filters by name or module while it reads and
answers at most 768 KB a page with capped, total and the cursor next, in
order of the names; container-image-registry's answers whole as its form
says. The registry takes a removal for retention whoever asks it: the
refusal "not the sweep's" and its read of the sweep seat's holder go. The
sweep reads the list from its cursor until no page is left.
Builds: design 47 H2 §813 (7), §811, §779; H-A-350 DMC-10, H-A-399
A 401, 502, 503 or connection fault is judged by the forge's own
unauthenticated version read: the read failing, or naming the store
unavailable, answers the kind away with part and words, which every caller
reads as a transport fault; the read passing, a 401 (the fresh token refused
too) is the forge refusing the mesh's admin account, never worded as an
account to create.
Builds: design 47 H2 §776; H-A-350 DMC-2, H-A-354
The module's start is the test's setup: its two waits on the bus take
testbus.SetupBound (half of MESH_TEST_LEVEL_BOUND) and say a starved setup
in the one line testbus.Starved writes; its 30 s values go.
Builds: design 47 H2 §800; H-A-327, H-A-368
tags answers items {tag, commit} read from the forge's own tag list, a tag
the forge names no commit for being an error, never a tag without one; tag
writes <directory>/vMAJOR.MINOR.PATCH, the form the Go toolchain's direct
fallback reads, go/v<version> among them.
Builds: design 47 H2 §796, §780; H-A-350 DMC-3; mesh-delivery D5 d2d3ebbf
gate_times answers per class its newest delivery (last) with whether it was
over its budget; the self-check's probe D-budget reads gate_times and raises
over-budget with the class as its id, naming the source revision and its
longest phase, and a newest delivery within its budget clears it.
Builds: design 47 H2 §810; H-A-396
The merge of main's H-A-387/H-A-393 fix gave Store.Held a reference and the
digest another publish holds; a repeat answers ended when the reference is set.
Builds: design 47 H2 §775; H-A-381
The head entry keeps merge-called from its opening and the pipeline's phases
from its end; a record whose every mover runs ends with delivery-time (from
merge-called to the last running of its movers) and class (core when a mover's
catalog entry, read once at its entry's opening, says core). gate_times answers
each class's budget (budget-core 10m, budget-leaf 3m), the last seven days'
median and worst, and every delivery over its budget with its longest phase.
Builds: design 47 H2 §810, §785; H-A-394
Each build module declares again-delays (15s,60s) and gives its puller the
delays steprun.AgainDelays reads from it; a publish's no responders is a
transport fault by steprun.PublishFault with no seat call to learn why, so
the modules no longer invoke the assigner's assignments.
Builds: design 47 H2 §799; H-A-350 DMC-7, H-A-390
The test bus serves the ledger's query the movers index answers, and the
take-in's mover is looked for in its own entry, the head counting it.
Builds: design 47 H2 §809; H-A-391
The gate reads the registrations page by page from the cursor each page
answers, so a source-only registration (no module) past the first page is
watched as any branch; its merge's take-in with no movers ends the record
nothing to deliver as before.
Builds: design 47 H2 §815, §806, §779
A record's head entry record/<branch.source> keeps the record-wide fields and
the count of its movers; each mover, the catalog half's too, has its entry
mover/<branch.source.module> with its pipeline, version and step outcomes per
node, found through the index movers on its record's key. An entry over
64 KiB is refused with record-too-large, cleared when written within it or
at the record's end; the ledger is capped at 1 GiB.
Builds: design 47 H2 §809, §820; H-A-385, H-A-391
While the reserved digest is still empty, a second publish of the same
version with another digest passed the reservation check and the store's
Put moved the version's tag. Publish now refuses it in its held branch,
before the reservation is asked, naming both digests, with the publish's
finding and refused said as any refusal; the same digest is taken as before.
The store's Held answers the digest it holds at the version when it differs.
Builds: design 47 H H-A-393, design 47 G A-490j
deliveries takes module, outcome, since and the cursor next beside state,
repository and group, applies them while it reads, and answers at most
768 KB a page with capped, total and the next page's opaque cursor.
Builds: design 47 H2 §811, §779
The builder records a build's digests only at its end, while the Go build
module publishes during the build step with the reserved version (H-A-373),
so the distribution holder's strict digest check refused every archive-only
publish with "reserved ... for another digest". Recorded now takes an
artifact whose reserved digest is empty, as A-490j decided for every holder,
still refuses a different recorded digest, and its refusal names the case:
the digest reserved and the one published, or no reservation of the version.
Builds: design 47 H H-A-387, design 47 G A-490j
A publish carrying an ask id is kept in the state taken for ask-taken-bound
(10m): a repeat while the act runs joins it, a repeat after it ended answers
the act's state read afresh (the transfer's lacking parts, a held version
ended, a transfer gone refused), never publishing twice.
Builds: design 47 H2 §775, §820; H-A-359, H-A-381
Every delete of a chunk in the incoming object store is followed by the
runtime's purge of what it left (busstate.Objects.Purge), so the store's kept
descriptions equal its live objects; a failed purge is said.
Builds: design 47 H2 §813 (8); H-A-227, H-A-366
Every bundle and archive was published with an empty version, a by-digest
publish, so the builder marked none against its reservation: the same kind
as the Node build module's bundles.
Builds: design 47 H H-A-373
Every bundle was published with an empty version, a by-digest publish, so
the builder never saw package-published and the bundle was not marked
against its reservation.
Builds: design 47 H H-A-372
The builder sends a module's directory with every lane, and an archive is
packed by the Go build seat, so the fetch's refusal of a directory without
go.mod left every archive-only module without Go code unbuildable. Such a
directory is now an empty scope (the empty scope file written); a Go bundle
or package without go.mod still fails at its build step, naming it.
Builds: design 47 H H-A-371
mesh-contracts' K1 regenerated every seat package to answer (value, form, error); the catalog's four
callers take the form.
Builds: design 47 H2 §780; H-A-350 DMC-4
The forge module needs a place for its token that only its own processes read: own-put keeps it as
module/<module>/<name>, named by the caller's account, its newest generation only.
Builds: design 47 H2 §822, §808; H-A-345, H-A-349 DQ-7, H-A-360
Since the toolchain lock pins no SDK (design 47 G §571), only the fetch puts
@novox/mesh-sdk in a module's node_modules. The three tests that build with
the real tsc and esbuild skip the fetch, so they read the SDK from an older
installed toolchain that still carried one. Once a build node installed the
current lock (shanks 19:26, then g14), they failed with TS2307 (main.137,
main.155). The fixture now lays a minimal SDK of its own where the fetch
would, so the tests no longer depend on which toolchains a node still keeps.
Builds: design 47 G §571
Review round 2 (1): the start's wait for the expiries the watch handed over
had no test that failed without it; the start test gains a store whose
deletes wait, and without the wait the key is said expired twice. The note:
a node-engine package naming no verb converged refuses the start loudly
instead of asking seat:node-engine.undefined.
Builds: design 33 control-verbs §10, §11, §14; design 47 H H-A-312
Review round 1 (1)-(3): an until without seconds was accepted, which the
node-engine's RFC 3339 read refuses, withholding every later compose; the
verbs answered before started was said and before the watch's expiries
ended; and the holder kept its own copies of the seat's names, key forms
and value forms, which had drifted from H-A-324. The holder now takes them
from the seat's generated package, the tools' inputs are checked against
its input types, and every seat verb waits for the start's end.
Builds: design 33 control-verbs §2, §3, §11, §12, §14; design 47 H H-A-324, H-A-330
An allowance with no interface never opens an outward link; one named in
interfaces does, and allow now says so: a kept allowance whose interface is
in the outward links of the converged answer asked for the call answers
with outward, that link. The holder reads no shape of the loaded table to
tell an outward link from the tunnel; the event carries no outward.
Builds: design 33 control-verbs §4, §14; design 47 H H-A-324, H-A-325
The holder's ask of the node-engine's converged takes the seat's name, the
verb's name and the answer's form from the node-engine seat's TypeScript
package, as the holder says its own event through its seat's package.
Builds: design 33 control-verbs §10, §12; design 47 H H-A-316, H-A-321
The security and seams reviews of the packet filter's control verbs: the
allowances state is one bucket per machine whose keys carry the node, so a
holder lists and writes its own node's only; allow and disallow refuse an
unproven caller, an adopted node and a converged that does not answer within
10 s, and are written in the hand-act log before the state is; allow refuses a
port the loaded mesh table accepts from given sources only unless a source is
given, and more than the setting allowances-max; allowed answers none on an
adopted node and asks converged again while none answered; rules marks loaded
only on an equal at and lists loaded allowance comments nobody keeps. The event
is said through the seat's generated package; the claim says it, so emits
stays without it.
Builds: design 33 control-verbs §1-§11, §14; ADR 0683; design 47 H H-A-314 (3), (5), (6), H-A-315 (1)-(3), (5), (7), H-A-316, H-A-317, H-A-318, H-A-319
The operator, 2026-11-21 20:35: a firewall module has tools to control the
firewall beside the provisioning from declarations. The holder keeps one
allowance per protocol, port and interface in its state allowances, says every
change as allowances-changed, removes an allowance at its until by a timer on
its own state (and every expired one at its start, before started), and rules
marks each allowance loaded by its comment in the mesh's table. It writes no
rule: the node-engine renders the allowances.
Builds: ADR 0683; design 33 control-verbs §1-§7, §9; design 47 H H-A-304..310, H-A-312 (2)
The Go integration level fetches the forge module's tree without node_modules and
without network, so TestTheSeatGitWholeOverARealBus failed every pipeline carrying
the forge. It now checks its needs first and skips by name, as the Node build
module's realMachine does; with its needs present it runs whole as before.
Issue 1960 (H1).
Builds: design 47 H H-A-294
The test switched the peer's uid in a plain variable the keeper's serving
goroutines read, a data race that fails it under -race; it is an atomic now.
A peer the keeper refuses could also get "broken pipe" instead of the
refusal, when the keeper closed before the request was written: the client
now reads the refusal that waits on the socket. A request larger than the
socket's buffer makes that case happen every time.
Builds: A-37 (only root opens the keeper's socket; a non-root peer is told so); design 47 H section 10 rows 165 and 167 (hq branch design/47-H1-answers-59, 81fa1ee4)
The latest release is placed whatever the range a package.json names
(ADR 0573 §2); one outside that range is now said in the journal, for
the SDK and every other mesh package, so a module built against a
release its range excludes is visible. A range of a form not read is
said as such.
Builds: design 47 H H-A-267, H-A-276
A fetch naming a member's build of a mesh package the registry lacks
lays the build and waits on nothing; without the skip in
placeMeshPackages it waits for the registry's release.
Builds: design 47 H H-A-267
The keeper never marks a first view listing no module as answered
(A-367), so the subtest waited for what cannot come and hid every later
subtest. It waits for the first fill pass and takes it as the first.
Builds: design 47 H H-A-268
A trunk build put only the SDK into a module's node_modules; any other
@novox/mesh-* package its package.json names was left out of the install
and only a group or a member's build brought it, so the forge module's
build failed with "Cannot find module" for its contracts package. Each
such package is now placed at the latest release the registry holds, as
the SDK is; a group's packed package or a member's build still takes
precedence.
Builds: design 47 H H-A-267
A done build whose history read failed was left out of the view for the whole process, so later passes judged from an
incomplete keep set, and every build cost one history read. A build's end is now taken from the build-ended entries of
the builder's log the pass already reads once, and only a build the log lacks is read from its history. While any done
key's end is unread, or the view is not yet taken in, no pass judges: it ends with the finding artifact-builds-unread,
written once and cleared by the next pass that judges. Every change heard while a watch's handover is gathered now goes
into that handover, so a build finished between the handover and its taking in is not lost by the swap (H-A-257).
Builds: design 47 H §H-A-255 (1), §H-A-257, §H-A-258, §H-A-260
A watch asked again after an error handed over the current values on top of the old view, so a key deleted while no
watch ran stayed in the keep set. Each handover is now gathered apart and replaces the view whole once it is in, and
the ends of keys it no longer holds are forgotten.
Builds: design 47 H §H-A-255 (1), §H-A-257
A page read costs the whole bucket per page, so the builder's builds are now a followed view: the watch's first view
at start, then its updates, kept in memory with each build's end read once from its history. A pass reads nothing of
the builds on the bus and judges nothing before the first view is in, so it never sweeps against an empty keep set.
Builds: design 47 H §H-A-250 (2), §H-A-255 (1)
A finished build or a want change started a sweep and an observation each, every one reading the builder's builds key
by key and asking the registry's list again, queued behind one lock: a third of the night's bus flood. Now an event
marks the one pending pass, a pass running when marked runs once more, the builds are read in pages of the state with
each build's end from the page, and a list the registry seats cannot answer ends the pass with artifact-list-unread,
written once and cleared by a pass that reads it.
Builds: design 47 H §H-A-250 (1), (2), (3)
The router held the bus module's destroy-level asks with the bare words
"no channel can carry it now", so the asker could not tell why. holdWhy
now names the destroy level and that no channel carries a destroy answer
yet. The bus module's tool says "offered, channel not yet named" for an
open record with no tickets, never "offered on :".
Builds: design 47 H H-A-237 (3), H-A-239 (2)
The bus module's drop and rebuild asks were at the destroy level, which no
channel carries yet, so the router held every one unshown until it expired
while the tool answered "the operator is asked". The ask is now at the
approve level for the caller's expires_minutes (5 to 1430, default 60,
refused outside), the keeper accepts an approve-level choice for these two
acts and bounds a request by its ask's own expiry, and the tool makes one
read of the router's record after the publish and says offered, held or
refused as it finds it; the request read shows that standing too.
Builds: design 47 H H-A-237 (1) (2), H-A-239 (1) (2)
A directory of a trees listing's files the commit does not hold was
thrown as a plain error and answered with the kind failed, so the
indexer took it for a forge that failed, never ran its one-directory
fallback (issue 1585), and published nothing of the repository.
commitTrees throws DirectoryNotInTree for it, and the seat's failure
answers it with the kind not-found, the error naming the directory.
Builds: design 47 H H-A-232 (2), H-A-234; issue 1585
Review round 1. keepOld took the old plan's objects whole, so a state's `<state>-objects` store
the grants part made beside a bucket the keeper made alone would have been compared, grown and,
on a recreate, deleted by the keeper: it now keeps only the old plan's objects the made record
already holds. CapSum skipped every kept entry as counted in the old sections' caps, which count
objects, logs and streams but never a state: it now skips only a kept objects, log or stream
entry, so a kept state's bucket stays in the sum.
At the delivery the plan digest changes once for every module with old sections (most modules,
since state counts): each plans again once, quietly, as its entries are compared with what the
bus holds and nothing is created or said where they agree.
Builds: design 47 H H-A-229, H-A-230, H-A-231; G §547
The keeper read only a manifest's new-form asks, so the entries it had made for the artifact
registry and the forge, now declared in their old sections, stood as orphans with live data: a
rebuild refused, a drop offered. The keeper now reads every manifest's old sections (state,
objects, logs, streams) by the SDK's naming per section, as the grants part composes them. An
entry it made under one of those bus names is no orphan, is compared and grown to the declared
cap and rebuilt on request; it creates none of them and says no created for one, and counts a
kept one's cap once, in the old sections' sum.
Builds: design 47 H H-A-229, H-A-230, H-A-231
A one-line comment in ~/.fehbg (the files archive) moves feh's fingerprint and
is readable on every machine that holds feh once it is delivered, with no hand act.
Design 47 H §5 runbook step 7: "Prove it. One single-module change (a comment in
a module with lint, unit and composition steps only), merged to main through a
merge call (merge-pull, design 47 H §734), followed from take-in.taken through
pipeline-ended to the node-engine's running and the gate's ended record, with no
hand act".
G §370: "The version reaches every artifact's build": the delivered files
archive carries the version the pipeline gave it.
Builds: design 47 H §5 step 7, G §370
A requirement of the checked repository at a real version was fetched from the Go seat while the
indexer's fingerprint takes the tree, so the build and its key disagreed; and "same repository" was
read from the owner name novox/ written into the code. The fetch now takes every requirement of its
own repository from the fetched tree before anything is downloaded, at whatever version go.mod
names, keeps those directories in the narrowed tree, and judges "same repository" by the fetch's
repository as the ask names it.
Builds: design 47 H S31, S3, L8 (H-pins-spec-and-gaps section 8 item 6); H-A-204, H-A-202, H-A-198
A package artifact's go.mod may require a package of its repository that
requires another of it at the zero version; the narrowing kept only the
first level, so the build of the second found no module of that path. The
entry's paths keep the manifest's form; the Go build module reads each kept
package's go.mod in turn, each directory once, and keeps what it requires.
Builds: design 47 H H-A-202, H-A-198
Given no pin, the module took the version of another package of the same
repository, or the latest, and built against versions never published. A
requirement of another repository is now taken at the pin the ask names for
its own path, else the build fails naming the path; a same-repository
requirement is still built from the fetched tree.
Builds: design 47 H H-A-192, H-A-198, G §370
disk-load's Go package diskreadings is imported by mesh-grants, whose build
waits for a publish no release made: an imported mesh package is a declared
artifact of its module, so every release publishes it. The gate imports the
assigner's asks, published in the assigner's series 0.3, so it names that
series in its packages.
Builds: design 47 H H-A-186, H-A-188, H-A-198, G §370
Review round 1: versionsAnswer's choice of digest per seat had no test.
TestTheVerbsSpeakTheSeatsContract now holds it, with a store whose blob
and manifest digests differ. OCI.Blob refuses a manifest that keeps no
single blob, naming its layers, and versions says that in the journal:
the contract's answer has no field for why.
Notes for H2: OCI.Held (oci.go) compares the tag's manifest digest with
a bundle's or an archive's blob digest, the same kind as this fault;
versions answers kind "" since the tool reads a kind its input does not
take.
Builds: design 47 G A-430a (G-22-lead-answers-2), design 51 *One contract*
The node-engine resolves a want's version to an archive's digest through
oci-artifact-registry's verb versions and fetches /blobs/<digest> by it.
The holder answered the tag's manifest digest, so shanks asked
claude-code/files' manifest as a blob and got 404 (claude-code 0.1.17).
On the artifact seat versions now answers the manifest's one blob, the
digest the reference and the event published carry; the image seat
keeps the manifest digest.
Builds: design 47 G A-430a (G-22-lead-answers-2), design 51 *One contract* (the seats' events carry the digest), the seat's contract Artifact and Listed.artifact
Review round 1. The node build module runs the seat generator's check at its unit level
(steprun.OutOfStep), so its narrowed workspace keeps what H-A-136 and H-A-137 name: every
generated-from directory of the module, and for a root module the seats/ directory and the
paths of every artifact naming generated-from. Narrowed bare, a step of one mesh-contracts
package named the generated files left out as out of step. The fetch and the steps verb now
narrow the same way, and the steps verb reads its narrowed tree as the levels do.
narrowTree and generatedFromDirs are copied from mesh-build-go without its Go part, not moved
into mesh-delivery-sdk's steprun: the move is a change to two more repositories and a group
delivery for some sixty lines, in a fix that unblocks the build steps now; the copy says so.
packageDir and narrowedAway share one path helper.
Builds: design 47 H H-A-136, H-A-137 (hq design/47-H1-answers-59), ADR 0499 clause 4
A step of one TypeScript package of a module with several (mesh-contracts' root:
ts-meshbroker beside ts-meshrecovery, pipeline main.3 of 2026-11-20) fetches a workspace
narrowed to that package's path. The fetch and the lint and test levels read every package
artifact of the manifest, and one whose path the narrowing left out fell back to the
module's directory, which holds no lock: every TypeScript step of mesh-contracts ended
"could not run: the package in . has no package-lock.json". Such a package is now left
out of the fetch's install and of the levels' targets; an unnarrowed tree is read as before.
Builds: design 47 H H-A-136 (a step's workspace holds the artifact's paths; nothing else is fetched), ADR 0499 clause 4
The licence manager says each binding change as binding.changed.<consumer>
(mesh.mod.claude-licence-manager.event.binding.changed.<consumer>), but both
manifests consumed claude-licence-manager.binding.changed, which the grants part
composes into the exact filter mesh.mod.claude-licence-manager.event.binding.changed:
the bus consumer never handed either module one change, and the vault kept only its
first view (2026-11-20: g14 refused a switched licence's token until the first view
was asked again by hand). The consumed pattern now names the further part, `*`.
Builds: design 47 G A-134; design 39 bus-and-switch §17, §19, §22
A login not handed is kept only against a lease that does not succeed it; a lease
of the same account that succeeds it replaces it, said as login superseded. A login
to another account is kept until the manager adopts it.
Builds: ADR 0206, ADR 0209, design 39 §6, H-A-138 (hq design/47-H1-answers-59 175ed320), H-A-139 (hq design/47-H1-answers-59 e359911e), H-A-140 (hq design/47-H1-answers-59 b75d1386)
The manager's refresh keeps the licence's refresh-token expiry, so a grant of the
same lineage with a later access expiry replaces the login; a grant with no
mark, an older lineage or no later access expiry is the older lease and is kept.
Builds: ADR 0206, ADR 0209, design 39 §3 §6, design 14, H-A-138 (hq design/47-H1-answers-59 175ed320), H-A-139 (hq design/47-H1-answers-59 e359911e)
One reason, login kept, on both paths, with a journal line; the not-handed
return skips the binding's generation on purpose.
Builds: ADR 0206, ADR 0209, design 39 §3 §6, design 14, H-A-138 (hq design/47-H1-answers-59 175ed320)
A login made on a node (while the tool runner was stopped, or before the
manager asked claude_code_grant for it) was overwritten by the next apply
of the binding's older grant: Apply took any refresh token in the file as
one already handed, and forced the write. The person's login was lost and
the session asked for a login again.
The node side now keeps the fingerprint of the login claude_code_grant last
sealed (state login-handed, never the token). Apply leaves a file whose
login is not that one untouched (applied false, reason login-not-adopted);
its report makes it the manager's candidate, and the successor the manager
hands back replaces it, as before without a refresh token.
Builds: ADR 0206, ADR 0209; design 39 §6 (the latest login wins, a login
moves its node); design 14 (a node never holds a refresh token).
With the guard gone nothing reads the machines' names, domains and
addresses: remove Machine, the name lookup and the ten-minute timer that
resolved them again, so the tools' table is read only when a followed
state changes. Also remove the now unused cut helper.
The operator asked for the shell guard to go (2026-11-20): the plugin no
longer carries a PreToolUse hook or guard.json, the binary no longer has
a guard verb, and the claude_code_guard tool, its record and the
refusal-ranked order of the instead-of table are gone. The SessionStart
and Stop hooks of the wait list stay, run by the same binary at the same
path, which is also the home-fs helper's. The rendered rule against ssh
to a mesh machine stays as the session's own and no longer claims a
guard enforces it; claude-agent no longer passes the guard's override.
A module at the repository's root naming generated-from "." (mesh-contracts)
kept nothing of it: the narrowing kept "." alone, which prunes everything,
so the seat generator's check of go-meshconfig still found every file out
of step. Such a module now keeps the seats/ directory the generator reads
and the paths of every artifact naming generated-from; nothing else of the
root.
Builds: design 47 H H-A-137
The first pipeline of the clean state (main.3) failed every build of a
mesh-delivery-contracts client: the fetch narrowed the tree to the
artifact's paths (ADR 0499 clause 4), which removed
modules/mesh-delivery-sdk, so its requirement at the zero version found
"no module of that path in the fetched tree". The same narrowing left the
seat generator's out-of-step check without its seat definitions.
The narrowed workspace now also holds every generated-from directory an
artifact of the module names and the directory of every package of the
same repository the artifact's go.mod requires at the zero version, found
by the repository's manifests; the root's manifest and lock files are not
kept when narrowing from the root. The fetch's own Go modules stay those
under the artifact's paths (A-447a), not the added directories.
Builds: design 47 H H-A-136; design 47 G-22 A-447a
Claude-Session: https://claude.ai/code/session_01JG73KHUnBze48rdua8u2sy
The refusal read the keeper's state asked, which holds only a module that
asks for kept data, so a running module keeping no bus data would lose its
live durables. It now reads the keeper's state plans, one key per module of
a watched branch's newest revision; the keeper no longer deletes a plan key
when a module's asked key goes, so plans stays that accepted set.
Builds: design 47 H H-A-126, H-A-127, H-A-128; design 47 G §538
Claude-Session: https://claude.ai/code/session_01JG73KHUnBze48rdua8u2sy
An accepted module's durable (<node>_<module>) is refused on any stream; any
other durable is an orphan and goes, why its warrant, recorded through
mesh-hand-acts. It no longer refuses every consumer of a kept-data stream.
Builds: design 47 H H-A-126, H-A-127
Claude-Session: https://claude.ai/code/session_01JG73KHUnBze48rdua8u2sy
Every open record ends after one hand act naming how many; the answer says
what it ended, counted.
Builds: design 47 H H-A-121 (2), §765 (the gate's open records go at the clean state)
Claude-Session: https://claude.ai/code/session_01JG73KHUnBze48rdua8u2sy
Every account may call every verb; what protects the mail is what the verbs
answer (ADR 0459 §2), not who may call them.
Builds: design 47 H §764 (every callers list, §752); H-A-122
Claude-Session: https://claude.ai/code/session_01JG73KHUnBze48rdua8u2sy
Review round 1, notes: keep-main-rules acts at the token's mint alone since the
ten-minute step went (H-A-100), and the issue tracker's owners named the
checker's path, which a search for the checker must not find (§764).
Builds: design 47 H §764 (the forge module's ten-minute timer; the checker whole); H-A-100
Claude-Session: https://claude.ai/code/session_01JG73KHUnBze48rdua8u2sy
Review round 1, Must 3: the groups and group-ended are retired; the overview
showed the deliveries, another module's state, which the status view owns.
The overview keeps two sources (the tracker, the asks), consumes no merge,
transition or pipeline's end, and the progress verb asks no groups.
Builds: design 47 H §764 (the gate's verb groups; the runner's group-ended); H-A-28; H-A-125
Claude-Session: https://claude.ai/code/session_01JG73KHUnBze48rdua8u2sy
Review round 1, Must 5: the seat lost held-back and ask-again (mesh-contracts
9db2e1d); the holder still offered held-back asks of before, refused choices
on them, listed them in asks and served ask-again.
Builds: design 47 H §764 (The operator channel's limit); H-A-57
Claude-Session: https://claude.ai/code/session_01JG73KHUnBze48rdua8u2sy
Review round 1, Must 2: the want's composed digest left mesh-sdk's motion.Want
(H-A-78); the hold read it and no longer built.
Builds: design 47 H H-A-78 (1)
Claude-Session: https://claude.ai/code/session_01JG73KHUnBze48rdua8u2sy
Review round 1, Must 1: commit 11a22fdf removed the checker's seat package;
the require left the gate unbuildable against the H1 heads.
Builds: design 47 H §764 (the checker whole goes); H-A-57
Claude-Session: https://claude.ai/code/session_01JG73KHUnBze48rdua8u2sy
A Go integration test builds the TypeScript bundle and plays its tool
runner over stdin and stdout against a bus of its own and a fixture
forge: a call stopped after two of three merges says one merge-called at
the start; a failed required step refuses its head; two pull requests in
two repositories merge and say one merge-called; a forge refusing the
second merge says what landed and raises merge-partial, cleared by a
later call.
Builds: design 47 H §774 (forge module: tests 111, 114, 115), §734, §735, §736, §766, §767; H-A-61, H-A-87 (1)
Claude-Session: https://claude.ai/code/session_01JG73KHUnBze48rdua8u2sy
announceHead with its mesh/merge-gate pending, the weighing of trunk and
sibling moves (passPulls, trunksMoved, siblingMoves, GateWriter) and their
files on disk, the pass's pull-passing half with its 60 s retry, and the
10-minute keep-main-rules step go. A pull request's new head is said as
pull.updated from the forge's event alone; the seat's status writes
through the client; pollRepos, pollMerged and readAllAtStart stay.
Builds: design 47 H §764 (the forge module: the merge-check machinery, its poll and timer), §762; H-A-28, H-A-100
Claude-Session: https://claude.ai/code/session_01JG73KHUnBze48rdua8u2sy
The forge module keeps a step's failure at the step-ended body's
fingerprint alone, consumes step-ended on its <module>.<branch>.<commit>
subject, and writes step-unread (scope module, keyed <run>.<module>.<step>,
warning, resolver operator) for a step-ended without its fingerprint,
cleared by the step's later step-ended with it.
Builds: design 47 H §735 (failed-steps from step-ended); G §585; H-A-18, H-A-47 (FORGE-1 builds step-unread), H-A-93
Claude-Session: https://claude.ai/code/session_01JG73KHUnBze48rdua8u2sy
forge_calls answers the kept calls newest first with each member's base,
head and state; forge_call_say_again, with call and why, says a call's
merge-called again once the hand-act log took it.
Builds: design 47 H §766 (its tools); H-A-18
Claude-Session: https://claude.ai/code/session_01JG73KHUnBze48rdua8u2sy
The call stops merging, says merge-called with what landed and refused,
and raises merge-partial (scope delivery, id the call, urgent, resolver
operator) in the module's findings; a later call landing every refused
member clears it, and forge_partial_clear clears one by hand after the
hand-act log took it.
Builds: design 47 H §767; H-A-1, H-A-18, H-A-81
Claude-Session: https://claude.ai/code/session_01JG73KHUnBze48rdua8u2sy
Once its members are merged or refused the call says merge-called on
merge-called.<call> with its landed members (base and merge commit) and
its refused ones, and is kept said; a call that landed none says none. At
the start every call not said says it for what landed.
Builds: design 47 H §736, §766 (said again at the start); H-A-18, H-A-47 (said), H-A-71 (2)
Claude-Session: https://claude.ai/code/session_01JG73KHUnBze48rdua8u2sy
The key call.<call> holds by, the proved account, at, each member with
its base and state (judged, then merged with its commit as it merges, or
refused with why) and said; a call past call-retention (a setting, 30
days) is deleted on a new call's write, oldest first, with no timer.
Builds: design 47 H §766 (the state calls, call-retention); H-A-18, H-A-47 (said, deletion on each new call's write), H-A-71 (2)
Claude-Session: https://claude.ai/code/session_01JG73KHUnBze48rdua8u2sy
merge-pull takes pulls and by, refuses a call with no proved caller,
judges each member (its head, its base a watched branch, not merged
already, no required step failed at its fingerprint) and no branch
protection, merges none when one is refused, else merges in the order
named and answers the call and each member's outcome once. The seat's
branch answers no protection. The seen-expiry test no longer depends on
today's date.
Builds: design 47 H §734, §735, §764 (the single-pull form; merge-pull's protection judgement); ADR 0682 §13; H-A-21, H-A-22, H-A-24, H-A-47
Claude-Session: https://claude.ai/code/session_01JG73KHUnBze48rdua8u2sy
The source key is read with the escaping it is written with, so a merge
call's head on a branch holding a dot is no untaken head. The whole test
raises head-untaken for a head held past the bound, takes a merge call on
release/1.2, holds a push no call delivered, and holds nothing after the seed.
Builds: design 47 H §774 (gate: tests 110, 111), §737, §740, §742, §768; H-A-87 (2)
Claude-Session: https://claude.ai/code/session_01JG73KHUnBze48rdua8u2sy
The record a merge call opened takes the runner's pipeline-started,
pipeline-ended, wait-held and wait-ended, the assigner's wanted and
pipeline-wanted and the node-engine seat's running, read by its key, and
ends on pipeline-wanted (running: every tier runs; tier-late: failed) or
failed on a pipeline that did not pass.
Builds: design 47 H §738 (G §141 to §147); H-A-29, H-A-75, H-A-89 (1) to (5)
Claude-Session: https://claude.ai/code/session_01JG73KHUnBze48rdua8u2sy
The repository's gofmt check refused heads.go, headtools.go and main.go
as written; a stale comment of the stalled conditions goes with them.
Builds: design 47 H §740, §742
Claude-Session: https://claude.ai/code/session_01JG73KHUnBze48rdua8u2sy
The three build modules no longer give the step runner a callers reader
of the seat list (steprun.SeatListCallers and Puller.Callers go): no seat
definition names callers and every account may call every verb.
Builds: design 47 H §752, §764 (every seat contract: the callers lists); H-A-41, H-A-66, H-A-71 (3)
Claude-Session: https://claude.ai/code/session_01JG73KHUnBze48rdua8u2sy
gate_heads reads the state heads with each head's since and finding;
gate_head_take, recorded in the hand-act log first, names the held head a
source revision as a merge call does, with a call of its own, its record
marked by hand with who and when and its member <repository>@<commit 12>,
and the finding clears.
Builds: design 47 H §742; H-A-24 (hand, by and the time in the trace), H-A-48 (a hand take's member <repository>@<commit 12>)
Claude-Session: https://claude.ai/code/session_01JG73KHUnBze48rdua8u2sy
The gate consumes the forge bridge's head-moved, keeps in its template
state heads each head source.<branch> does not name, and raises
head-untaken (scope delivery, id <repository>.<branch>, warning, the
operator its resolver) once untaken-bound passes, on its one deadline
timer; a later source revision naming the head lets it go and clears the
finding. The stalled conditions, their state and the operator-channel
use go: the watchman's record-open is the one signal.
Builds: design 47 H §740; H-A-27 (head-moved's form, forgemsg), H-A-48 (the stalled conditions go; untaken-bound on the gate's one timer), H-A-81 (warning, resolver operator, no owner)
Claude-Session: https://claude.ai/code/session_01JG73KHUnBze48rdua8u2sy
For every watched branch the registrations name, one source revision at
each registered repository's head read through git.branch, start true, a
call of its own, source-moved said and no record opened; recorded in the
hand-act log first, the counts answered. It refuses to run while a key of
the earlier form is held, naming it (the source bucket is dropped first).
Builds: design 47 H §768; H-A-15, H-A-47 (the seed opens no record; the start pass writes through its path), H-A-76 (Q-G2: numbers from 1, refuses an earlier-form key)
Claude-Session: https://claude.ai/code/session_01JG73KHUnBze48rdua8u2sy
The gate consumes the seat git's merge-called and, per watched branch its
members name, writes source.<branch> once by compare-and-set keyed by the
call, numbered one above the branch's newest; opens that revision's record
keyed <branch>.<source> with the call, by and account in its trace; says
record-opened on record-opened.<call> and source-moved on
source-moved.<branch>.<source>. The take-in.taken writes its movers into
the record read by its key, and a take-in with no mover ends it as nothing
to deliver. The verb groups, the record's pins and pins-stated, forge-made
and the emitted transition go. One commit for both subtasks: the record's
opening is the handler's second half and neither compiles alone against
the contracts.
Builds: design 47 H §737, §738 (a take-in with no movers ends its record), §749 (the pins are the indexer's), §764 (the gate: the source moved on take-in.taken, the verb groups); H-A-13, H-A-19, H-A-24, H-A-25, H-A-28, H-A-29, H-A-33, H-A-47 (the start writes through the seed's path, no record), H-A-48 (pins, pins-stated, transition, members <repository>@<number>), H-A-76 (Q-G1: the say-again carries the value's heads, start false)
Claude-Session: https://claude.ai/code/session_01JG73KHUnBze48rdua8u2sy
Every account may call every verb from H1, so secret-ask answers every caller the runtime
proved and retire every caller whose warrant holds; the manifest's two lists, the config
retire-callers and its file go. The enrolment verbs keep their one list, now a small type
of the vault's own since mesh-sdk's Callers goes.
Builds: design 47 H §752 (ADR 0682 §10's exception for the enrolment caller, H-A-11); H-A-41, H-A-66
Claude-Session: https://claude.ai/code/session_01JG73KHUnBze48rdua8u2sy
mesh-sdk's callers set goes in H1, so the seat list's two tools of claude-code answer
every caller, and mesh-recovery's seat list part is made with its prefix alone.
Builds: design 47 H §752, §764 (every seat contract's callers); H-A-41, H-A-66
Claude-Session: https://claude.ai/code/session_01JG73KHUnBze48rdua8u2sy
#708's six commits are taken onto main (the pull request path and A-490l's sweep go).
manifest-accepted goes in H1, so a retiring value is judged for every module a
catalog-revision names changed, and the resync tool's words no longer name the
indexer's config candidates, which retire too.
Builds: design 47 H §763 (G §654), §764 (mesh-config: its pull request path and A-490l's pull sweep; the indexer's manifest-accepted and config candidates); H-A-49, H-A-61
Claude-Session: https://claude.ai/code/session_01JG73KHUnBze48rdua8u2sy
The timer was rearmed on a write of proposals or retiring only, so a rename window opened after start, with nothing else
waiting on a time, left it blocked and its seven-day held finding never came; a window opened under a later armed deadline
came late. The condition is now armsDeadline (proposals, retiring, renames), and a test shows a written window rearms it.
The comment on Proposal.Pull cites ADR 0681 §5.
Builds: ADR 0681 §1; lead answer A-490l (hq main, 03-DESIGN/01-to-be/47-milestones/G-22-lead-answers-2.md: the held finding comes in a quiet mesh too).
Claude-Session: https://claude.ai/code/session_01JG73KHUnBze48rdua8u2sy
The one timer left is set for the earliest of three deadlines: a waiting proposal's end, an unanswered retirement's second
ask, and a rename window's seven-day held finding, which is now judged at its deadline (JudgeHeldWindows) so it comes in a
quiet mesh too. Its floor is the module's setting deadline-floor-seconds (default 60), not a constant.
Tests: an integration test counts asks by key and fails on any `seat:git.*` over the whole run (ADR 0681, How it is checked,
clause 1); a unit test shows an approved proposal records a hand act naming no pull request; another that a stored proposal
naming a pull request is refused on approval, saying why.
Claude-Session: https://claude.ai/code/session_01JG73KHUnBze48rdua8u2sy
The generated client held only the held values' keys and decoder, which the seat no longer has.
Builds: ADR 0681 §1, §6 (the seat keeps no held value, so its generated client goes).
Claude-Session: https://claude.ai/code/session_01JG73KHUnBze48rdua8u2sy
The mesh knows no pull request (design 47 G, G-1 §38-39). Removed:
- the states `pulls` and `kept` and all code that reads or writes them (pulls.go, the held values a proposal holds, the
listing of held values), SweepPulls, settlePull, HeardPull and the forge call;
- the consumes of git.pull-merged and git.pull-closed, the invoke of seat:git.pull;
- the config verb's `pull`, `head`, `declares` and `pulls` arguments and its actions `pull-declares` and `held`;
- the indexer's config candidates, which only that path used: the consume of mesh-delivery-indexer.views.config-candidates
and the invoke of indexer_read_config_candidates (the indexer's state goes in the same group).
A stored proposal that names a pull request is judged not one this module makes, so nothing is set on it.
Builds: ADR 0681 §1 (no pulls or kept, no git events, no seat:git.pull, no `pull` argument, no pull-declares or held), §3, §5, §6;
lead answer A-490l (a stored proposal naming a pull request is refused on its approval).
Claude-Session: https://claude.ai/code/session_01JG73KHUnBze48rdua8u2sy
mesh-config sent about 262 requests a second and received about 1,540 since its start: every sources() call (a
resolution pass, and a `delivered`, `manifest-reads-changed` or `manifest-accepted` event, each of which read the sources
whether or not a window or a retirement was open) read every key of its own states by one get each (pulls 1,176 keys,
resolved 447, declared 168, values 101...: about 1,200 requests for an event, 2,800 for a pass), serialised by one lock, so
a delivery's burst of events kept it asking the bus back to back.
- The module's own states are read from mesh-sdk's WatchedStates (one watch per state, its own writes taken into the copy)
instead of a get per key (ownstates.go, main.go runtimeStore).
- An event judges what it names: `delivered` and `manifest-reads-changed` read no source when no rename window is open,
`manifest-accepted` reads none when every row is already marked or read again (renames.go windowOpen, retiring.go).
- Open windows and retirements are judged again after each pass, since an entry that answers after the judgement was
made moves the pass, not the event; the slow reads had hidden that race.
- No timer polls: the five-minute resolution and sweep of every pull request on the forge, and the minute's read of the
router's record, are gone. The sweep and Decide run once at start (an event said while the module was down is not said
again), then on git.pull-merged/-closed and operator-channel.decided.mesh-config; an ask ended with no choice also
takes the retirements from the router's record. One timer is left, a deadline: for the earliest waiting proposal
(ProposalLasts + 1 h) or unanswered retirement (HeldFinding), set again when either is written (NextDeadline).
Wake-ups kept, each by its record: `delivered` and `manifest-reads-changed` close a rename window (design 32 *A config
value renamed*, ADR 0609 §4); `manifest-accepted` judges one module's retirements (design 32 *A value no longer read*);
`git.pull-merged` and `-closed` settle one pull request's held values (ADR 0502 §7); `decided.mesh-config` takes one ask's
warrant (ADR 0259); a changed view (assignment, want, predefined list, config candidate, catalog revision, entry) moves a
resolution that writes only what moved (ADR 0565 §3).
Builds: ADR 0681 §1 (no polling of the forge: the five-minute look ends); ADR 0580 §1; ADR 0426 §3; design 32 (*A config value renamed*, *A value no longer read*);
lead answer A-490l (hq main, 03-DESIGN/01-to-be/47-milestones/G-22-lead-answers-2.md) for the wake-ups and the one deadline timer.
Claude-Session: https://claude.ai/code/session_01JG73KHUnBze48rdua8u2sy
From H1 only a merge call names a source revision, so the gate stops moving source on
take-in.taken and on the forge bridge's merged message, and the start's comparison with
the forge's heads, gate_source_follow, its finding source-uncompared and the forge-made
record opener nobody calls go with them.
Builds: design 47 H §737 (moveSource leaves the take-in.taken handler; ForgeMerged, moveHead, gate_source_follow go), §764 (the gate: the source per merged message, the start's comparison with the forge's heads, the source moved on take-in.taken; forge-made per H-A-25)
Claude-Session: https://claude.ai/code/session_01JG73KHUnBze48rdua8u2sy
A merge call is the delivery's group from H1, so the gate no longer keeps deliveries
and groups from the seat git's pull events, reads no open pull request at its start,
and answers deliveries and show from the records alone. The verb groups stays,
answered from the records, until the seat mesh-delivery's definition drops it.
Builds: design 47 H §764 (the gate: the groups from a branch name, its states deliveries and groups, its consumption of pull-updated, pull-merged and pull-closed, its start reconciliation); H-A-28
Claude-Session: https://claude.ai/code/session_01JG73KHUnBze48rdua8u2sy
The switch to the records ran once at G's delivery; from H1 one model holds, so the
switch, the carried Approve and its finding, the states merges and the old owed, the
ledger's switch key and the checker client nobody imports leave the gate.
Builds: design 47 H §764 (the gate: the switch and the carried approval, the states merges, owed, asks, pulls, turns); H-A-28
Claude-Session: https://claude.ai/code/session_01JG73KHUnBze48rdua8u2sy
The builder's publish again sends version and digest and no parts; the package seats refused it as not one part.
Both registry holders now run the publish conformance cases.
Builds: design 47 G A-73, A-417, A-490
Claude-Session: https://claude.ai/code/session_01JG73KHUnBze48rdua8u2sy
mesh-build-image's steps now carry Node and SayStep (an image has no step, §328, but the kit holds all three). Prior wired in go and image.
Builds: design 47 G §24, §328
Claude-Session: https://claude.ai/code/session_01JG73KHUnBze48rdua8u2sy
A pipeline step ask (the sdk's lib lint and unit steps) was refused with 'this build module says no step's end': its
AskSteps set no SayStep, Node, Lint or Has. They are set as mesh-build-go sets them; lint runs tsc --noEmit over each TypeScript
package; Puller.Prior counts only runs that started toward poison.
Builds: design 47 G §328 (Lint per language), §22, §24, §25
Claude-Session: https://claude.ai/code/session_01JG73KHUnBze48rdua8u2sy
mesh-build-go and mesh-build-image follow the seat list as mesh-build-node does, so each consumes mesh-catalog.seat-definition-changed and
-removed, and all three invoke seat:mesh-catalog.seat-definitions, which the reader calls.
Builds: design 47 G A-488a, A-488b (hq 57c14047)
Claude-Session: https://claude.ai/code/session_01JG73KHUnBze48rdua8u2sy
Its runtime follows the seat list (seatlist.Start subscribes to mesh-catalog.seat-definition-changed and -removed); with no consumes
there was no consumer shanks_mesh-build-node_tools, the seat list was never read and the asks of a named asker were given back.
Builds: design 47 G A-488a (hq 25df48ad)
Claude-Session: https://claude.ai/code/session_01JG73KHUnBze48rdua8u2sy
Its processes run as the operator account, so its own secrets are owned by it (design 20 "Whose a process's own secrets are").
Builds: ADR 0680 §1, §4; design 20 "Whose a process's own secrets are"
The single-module proof of design 47 G (the operator, 2026-11-17), after pack49:
one change to one leaf module, followed to delivered on g14 and shanks with no hand act.
The single-module proof of design 47 G (the operator, 2026-11-17): one change to one module,
followed through its pipeline to delivered on g14 and shanks.
The toolchain a build module reports (guard key and toolchain-reported event) names the
language version alone (mesh-delivery-sdk steprun Guard); the build module's own
Toolchain func keeps the installed text. Delivered with or after the sdk change.
Builds: design 47 G A-471 (G-22-lead-answers-2.md)
The level took 67 s locally and 74 s in the pipeline, over IntegrationRun's 60 s. The eight slowest tests share no state and run in
parallel: 31 s locally with the fixture's store and bus. The real-store test makes its work directory under the test temp
directory, not /tmp.
Builds: ADR 0445 §3 (the integration level's 60 s bound)
A pipeline's test step places no repository beside the tree and a module's tests read nothing outside their module (lead answer
A-474), so the test no longer needs MESH_CHECK_BESIDE or the node-engine's tree: it fetches a small program of the composer's
contract (testdata/composer) and still proves the build once, the kept composer on the second ask, and the verdict by case.
Builds: design 47 G lead answer A-474 (principles 10 and 9)
The bridge kept a directory for the pre-G indexer's package publishing; its own commit says it goes once G's indexer runs, and
the test of the delivery SDK's step code refuses it. The unit level took 6.3 s over its bound of 5 s in the take-in's pipeline:
the digest tests ran their go commands one after another; the marker table is a subtest per case in a fixture of its own, the
list half a test of its own, and the six slowest independent tests run in parallel (4.6 s locally).
Builds: design 47 G §464, §680 (the build modules keep no copy of the delivery SDK step code); ADR 0445 §3 (the unit level's 5 s bound)
The forge bridge's merged moves the repository's head and says source-moved; at its start each head is compared with the forge's branch head; gate_source_follow does it for one repository by hand.
Builds: design 47 G-22 A-464, A-464b
bundleKeepDays read the config value snapshot-keep-days while allTools
listed the tools, before stdio.Serve; a config read waits for the runtime's
channel, which opens only once the bundle serves, so the bundle never
answered initialize and node-tools served none of its tools (novox
2026-11-16 22:21 and 22:26, after the first build from main since A-112).
The setting is now read when nats_snapshots is called.
A build seat's work queue (SEAT_NODE_BUILD_GO) held about 370 asks of
pipelines nobody waits for, ahead of the one pipeline wanted, and no tool
could empty it short of removing the stream. nats_stream_purge removes the
messages below a given sequence, keeping the rest, the stream and its
consumers; it needs why, reads first, offers dry_run, refuses a purge past
the stream's end and the module-kept data nats_drop owns, and says
stream-purged as its audit event, like the three removal tools.
Kind: a version written into a workspace's modfile without what that version's graph needs, judged against files the build does not use.
Builds: design 47 G (ADR 0573 §2, §4, as written in versions.go: asked versions hold whatever go.mod names)
A-379 replaced A-376 for the start, and the seat mesh-catalog offers neither the states modules and manifests as views
nor verbs of those names, so the validator refused the manifest's two consumes and two invokes.
Builds: design 47 G A-379, A-445 (G-22-lead-answers-2.md)
The operator channel kept at most three asks of one asker offered and held its oldest back behind newer ones, so asks
waited unseen (the pipeline's proposal behind two mailu asks). It keeps no count any more and holds nothing back: every
ask is offered when it is raised, and lapses at its expiry or ends at its answer. An ask held back before the change is
offered again at the router's next tick; ask-again offers one so and holds nothing back in its place.
Builds: design 47 G A-424
The builder's publish again of mesh-delivery-pipeline 0.1.6 sends the version and the digest with no parts, as A-73
has it; the build stored the bundle untagged, and the holder answered 'publish: no parts' unless the tag existed
already. The holder now checks the digest against the store under the name (the artifact store, A-50), asks the
builder's reservation, tags it with the version and says published; a digest it holds none of is refused as not
stored, naming the digest.
Builds: design 47 G A-417, A-73, A-50, §588
The forge's git verbs may serve after the gate does; one pass with every head read refused wrote nothing and was never retried.
Builds: design 47 G A-401
At its start and on each registration-changed of the seat mesh-catalog, the gate writes source.<branch> for every watched
branch the registrations name that no source key holds, each registered repository at its head from the seat git's verb
branch, group empty, then says source-moved; a repository whose head cannot be read is left out and said; a held key is
never rewritten. gate_source_start (branch, why) does it by hand through the hand-act log. The manifest consumes
registration-changed and invokes seat:git.branch and seat:mesh-catalog.registrations.
Builds: design 47 G A-401, §430, §15, A-34
The seed and a registration with no branches read the default branch through the seat git (design 47 G-1 §104) by
calling branch with an empty branch; gitea refused it as no branch for a path, and the seed stopped on novox 2026-11-14.
Live on shanks the bundle exited 39 times on 'an ordered consumer of mesh-conditions.findings-asked: context deadline exceeded', so the seat node-build-go never answered steps and the seed waited.
Claude-Session: https://claude.ai/code/session_01NcbgfGKaxw984XyAPfLm4h
Before the catalog half's first views are in (a read refused, or none-yet), the keeper reads the manifest on the trunk of the
holder of the seat mesh-catalog from the seat's states modules and manifests (the holder from the seat that places modules)
and makes its kept data as any module's, each entry said as ever; every other module waits as A-367 says.
Builds: design 47 G A-376 (G-22-lead-answers-2.md)
Claude-Session: https://claude.ai/code/session_01NcbgfGKaxw984XyAPfLm4h
PlanCatalog plans nothing, deletes no plans key and orphans no entry while the
catalog half lists no module before any was seen; the first revision with
entries plans as before.
Builds: design 47 G A-367 (G-22-lead-answers-2.md), W22/Q4 of start-waits.md
Claude-Session: https://claude.ai/code/session_01NcbgfGKaxw984XyAPfLm4h
mesh/check checking a4bc3d95edef on b1125c02cfe6 | novox/mesh-catalog#678 in group fix/47-G-bridge
The running pre-G indexer publishes mesh-catalog's packages only while every package directory it published before is
in the tree; G removed this one, so mesh-catalog's packages are never published (2026-11-13). An empty package keeps
the directory; it goes once G's indexer runs.
Claude-Session: https://claude.ai/code/session_01NcbgfGKaxw984XyAPfLm4h
mesh/check failed (2): mesh-delivery-checker: no published of the tagged version of go-meshconditions of me…; and 1 more | 53351dd30df2 on b1125c02cfe6
mesh-conditions keeps G's form: the keeper reads no observer's state (no reads, no Observers list), so main's zfs entry there goes. The zfs follower says its findings through G's findings writer over its keeper: the keeper gains the publish of the writer's three events on the finding stream finding-events and the hear of mesh-conditions.findings-asked on its own durable, nothing else; each pool or device not ONLINE is scope module, id zfs.<pool>.<device> (the device's own name), kind not-online, on the node. The manifest asks its state findings by template in requires (G §529, §577), and gains the finding-events stream and the consumes, and the keeper's bus the three events and the consumes.
Builds: design 47 G §383, §384, §387, §529, §577, A-339, A-340, A-341, A-342; design 42 zfs-owns-zfs §27, A-32, A-37, A-39, A-41
Claude-Session: https://claude.ai/code/session_01NcbgfGKaxw984XyAPfLm4h
mesh/check being carried to 0041b69d1736 on 32f1566cc879 | from 117f933dacf0, mesh-delivery 7221a1e5cd7f, mesh-host 0193949efe9f, mesh-sdk e42b9491c93…
The validator refused zfs and mesh-conditions: mesh-conditions reads zfs.findings, which zfs had moved to bus state in requires, where the new form has no readers (ADR 0596 section 2). The keeper process's bus.state keeps naming findings. The manifest test holds it.
Builds: design 42 zfs-owns-zfs A-46 (A-32, 26, 27)
In the test's temporary directory, or /tmp when the socket path would pass 100 characters, so a long TMPDIR cannot fail the bind.
Builds: design 42 zfs-owns-zfs §18, A-37
Issue 1979, blocking: a date-bound test. The warrant dated 2026-10-29 fell out of the listing's 14 days on 2026-11-13, failing every mesh-catalog check.
Builds: design 42 zfs-owns-zfs (the group's check)
The group's check failed TestRulesBus/NoCatalogueManifestNamesAnInstallation: 11 values name an installation (the vendor sites in the package's zed-functions.sh, resources[13]).
Builds: design 42 zfs-owns-zfs §14, A-2
The group's check failed TestOnlyRootOpensTheKeepersSocketAndANeedlessKeeperIsTheBusDown and TestAKeeperThatDoesNotAnswerIsWarnedOnce: stat /var/tmp: no such file or directory.
Builds: design 42 zfs-owns-zfs §18, A-37
The group's check failed TestEveryControllerKindHasItsHome: the reader zfs's findings are not read by the keeper.
Builds: design 42 zfs-owns-zfs §27, A-32
mesh/delivery-group group feature/1979-zfs-owns-zfs rejected: composed together: the checker's verdict at the members' heads (03f2ebff6ed0): unit: mesh-conditi…
mesh/check failed (9): unit: mesh-conditions/cmd/mesh-conditions.TestEveryControllerKindHasItsHome: the rea…; and 8 more | d15502f3650d on 117f933dacf0
Builds: design 42 zfs-owns-zfs §13, §24, lead answers A-45, A-43
The tools say every event of §25 (snapshot.taken.person, scrub.requested and the others) with mesh/stream.publish on the
stream events, never stdio.Emit; the hand acts' 30 s wait reads the mark (stream.fill) before the act starts, then
stream.read from the next sequence, for at most HandWait.
Builds: design 42 zfs-owns-zfs §25, §24, lead answers A-42, A-12
A failed keeper call in a root process (hand acts, forget, Paused, every call) now adds one warning that the bus is
taken as down, instead of being dropped silently. RunKeeper closes the connection acct.Dial opens when the keeper
stops; the message id's comment says it is nanotime-pid.
Builds: design 42 zfs-owns-zfs A-37
The three root processes declare no bus and reach the keeper's socket; the module invokes the service manager's start. The judge refuses the keeper's bus emits and state until the validator accepts what the module asks for in requires (A-38).
Builds: design 42 zfs-owns-zfs §25, §26, §27, §28, A-6, A-19, A-36, A-37, A-38
Renamed TestTheBundleRunsOnlyTheFormsOfADR0679: each root form is held to its whole argument list, no verb the decision never allows, zgenhostid only with the parsed hostid, a device only as cache or log.
Builds: design 42 zfs-owns-zfs §19
The manifest declares zfs-utils and zfs-dkms, renames scrub-pool to pool, and declares every new value not managed until set; its config file is ${config-json}, the options file, ZED's folder, rc, functions, handlers and drop-in each under their condition, the three ZFS units, the apply (run-once, restarted by its config file) and the run before the distribution's scrub and trim timers. One commit: the three subtasks change the one file module.json.
Builds: design 42 zfs-owns-zfs §6, §7, §8, §9, §11, §12, §13, §14, §15, §18, §20, A-1, A-2, A-3, A-7, A-14, A-15, A-16, A-23
zfs_config also says what the module says before it leaves a node (Z17): the next load of the kernel module uses ZFS's defaults, the three units and /etc/hostid stay.
Builds: design 42 zfs-owns-zfs §28, §30, §10, A-31, A-35
A trim schedule of weekly or monthly with no trim-day or trim-time is one refused value naming it; a keep count of 0 is that kind's none: the existing scheduled ZFS snapshots are kept and a warning names them.
Builds: design 42 zfs-owns-zfs §20, §21, §22, A-33, A-35
A value of another form is refused before anything is written or run; absent and empty are not managed.
Builds: design 42 zfs-owns-zfs §9, §11, A-15, A-23
The node-engine writes a cron step (*/15) as a calendar systemd refuses, so
zfs-mesh-scrub.timer never started and the pool had no scrub schedule. A list
of minutes is written as *-*-* *:0,15,30,45:00, which systemd takes.
Builds: design 42 zfs-owns-zfs §18. Tests: 1979/AC1.
The validator refused the manifest: a bundle's env may not set what the runtime gives every tool bundle. The processes
night and awake keep theirs.
Builds: ADR 0192
Claude-Session: https://claude.ai/code/session_01NcbgfGKaxw984XyAPfLm4h
intel-ucode failed an apply on a stale package database and stopped the delivery; an Intel machine
fails the unit by the image check and gets intel-ucode in the change that adds it. Tests: 1977/AC1
and the pinned order string adjusted.
Builds: L10 (issue 1977 note 6)
Before mkinitcpio -P, a preset naming ALL_config="/etc/mkinitcpio.conf" is copied to the state's
taken-over/ and loses that line (another ALL_config is left and named); after it, an image without
kernel/x86/microcode/ in its first 64 KiB fails the unit and records nothing. The unit may write
/etc/mkinitcpio.d. The README names no node. Tests: 1977/AC7, 1977/AC8; AC3/AC4 fixtures write images.
Builds: L7, L8, L9 (issue 1977 note 4)
Replaces the state record read (root's 0700) with images_with_microcode per /boot/initramfs-*.img:
true when the first 64 KiB hold a path under kernel/x86/microcode/; an unreadable image goes to
image_errors with its error. Test: 1977/AC5.
Builds: L6 (issue 1977 note 3)
Declares amd-ucode and intel-ucode, the drop-in /etc/mkinitcpio.conf.d/mesh-microcode.conf, the
initramfs script and mesh-kernel-initramfs.service; kernel_boot answers the running revision and
whether the images were built with the drop-in. Tests: 1977/AC1-AC5.
Builds: L1, L2, L3, L4, L5 (issue 1977 note 2)
mesh/check checking d22a7f3a412f on a60728e6f09d | novox/mesh-catalog#671 in group fix/1748-resolver-wait
The resolver file of systemd-resolved (resolvers and kept facts) carries options timeout:3 attempts:2 edns0 in place of timeout:1; manifest test asserts both facts, guard test's literal follows.
Builds: design 50 resolver-wait §1, §4 (and §2 of the main file, line 73)
Since A-290 a run whose program exited other than 0 ends its key failed, the
reason saying the exit, its answer kept (steprun.finish); the test still
wanted passed. It now wants failed with "the program exited 7" and reads the
kept answer as before.
Builds: design 47 G A-290
Claude-Session: https://claude.ai/code/session_01NcbgfGKaxw984XyAPfLm4h
Every TypeScript module of a fetch gets the registry's latest SDK (ADR 0573 §2,
A-284), so the harness's machine named the real registry and 11 tests (9 unit,
11 integration runs) reached the network; with the resolver down they failed.
The harness now serves a registry of the test's own holding one SDK release,
and the userconfig test reads its address from it.
Builds: design 47 G A-284, A-285
Claude-Session: https://claude.ai/code/session_01NcbgfGKaxw984XyAPfLm4h
Store.ListReadable lists every record and, beside them, each file it cannot read
by name with its error; RecordsHeld lists those as rows with an error and goes
on. recordsError stays for a store that cannot be listed at all. Store.List and
ListPrefix keep their behaviour for their other callers.
Builds: design 39 bus-and-switch §20; design 39 parts §12. Lead answers (issue 1409): V9 (keep recordsCount and recordsError), V10.
S1 vault-licence-secrets-listed. secret_holders answers `records` and
`recordsCount` beside holders, count and enrolments: each store record's name,
generation, since, changed, changedBy/changedVia, origin, length and the vault's
own fingerprint of the whole value, plus key and part for licence.<key>.<part>.
A record the vault cannot open stays listed with an error. Never a value.
Builds: design 39 bus-and-switch §20, §15; design 39 parts §12 (D1); design 24 Every secret lives in the vault, Beyond generate and hold; as-is 06 The vault, as it runs.
Lead answers (issue 1409 note 11): V1, V-thorough, V2-V8. V3 drops the map's
tokenFingerprint and AC5; V-thorough widens the licence.-prefix list to every record.
Both holders' uplink-tools copies read address, master, link and
link_index/link_netnsid from the existing ip -j address show call. Fields
are left out where a link has none (L3); a parent in another namespace is
{index, namespace} (L2); the form lives in uplink.go only (L4); only the
tool description changes, not seat.json (L5); bridge fdb MACs are not
answered (L6); names mac, master, parent (L1).
Builds: design 48 10, design 08 code:, issue 1965 note 11 L1-L6
Brings #664 (licences_switches carried entries), #665 (mesh-files lists a
place a secret is kept), #666 (provisioner: bounded retiring ask id, a
refused ask said, provisioner_health), #667 (mesh-build-go's step end
rewritten without the taken versions when over the steps state's bound),
#668 (mesh-files find writes the NUL itself). #667's buildstep/step.go and
digestshape_test.go conflict with G's deletion (the step runner moved to
mesh-delivery-sdk/steprun, SD5): G's side is kept; its intent, an answer over
the bound not lost, is G's A-290 (the answer kept as an output object).
Claude-Session: https://claude.ai/code/session_01NcbgfGKaxw984XyAPfLm4h
The keeper's hand-made receive (subscribe to the call's chunks, ask, receive in order, acknowledge each) moves into
mesh-sdk's process.Conn.CallChunks; the keeper's answer checks (an answered error, the space it fits, no chunk granted)
run as its accept hook before a chunk is taken.
Builds: design 47 G A-291
Claude-Session: https://claude.ai/code/session_01NcbgfGKaxw984XyAPfLm4h
Live on novox 2026-11-12: node-files.list of a root-only directory answered
'fork/exec /usr/bin/sudo: invalid argument'. The kernel refuses a program
argument that holds a NUL byte, and listAsRoot's -printf format and
searchBelowAsRoot's ended in a literal \x00 since 5d98bcc0 (2026-10-16), so
no list or search through root ever ran on a node. find writes \0 itself.
The stand-in root in the tests refuses an argument holding a NUL, as exec
does, so every root call is held to it; it found the search's instance.
Builds: design 39 parts §15 and the lead's answer D3 (list through root of a place a secret is kept).
A catalog check's answer of 264712 bytes was refused by the steps state
(256 KiB per value) and its end written failed without it, so the checker
lost the biggest member's answer and could not preview it. When a passed
or failed end is refused as too large, finish writes it once more without
the fetch's taken versions (fetch.packages, which no reader judges), the
step's reason saying how many were left out, why, and that the module's
journal names each; still too large, it ends failed as before.
Quick fix for issue 1921 (note 91); the thorough fix is design 47 G A-290
(an answer over the bound kept as the object output.<ask>).
Claude-Session: https://claude.ai/code/session_01NcbgfGKaxw984XyAPfLm4h
mesh/check checking 3bb8ace833d1 on c0d666fdcf55 | novox/mesh-catalog#666 in group fix/1966-unpublished-placed-module
The retiring ask's id is made from the key's digest and the time (at most 64 characters),
so a long key is asked. An ask the router's rules refuse is said in the key's why and in
provisioner_health. The Composition comment says what the code does: an unpublished
module keeps only its own plan and keys.
Builds: design 47 fix-unpublished-placed-module U6, U7, ADR 0593 §8
Issue 1409. readTools no longer appends the refusal sentence to list. A row in
TestSecretPlacesAreRefusedByNameBeforeAnyRead holds a link's target in a secret place to the content rule.
S2's match on sudo's English message is left as it is: the root path sets no locale, and sudo's environment
handling decides whether LC_ALL reaches stat.
Builds: design 39 parts D3 (§15), bullets 2 and 4
Main brought #663 (fff3b819) in mesh-build-go/buildstep/askrun.go; G moved that step
runner into mesh-delivery-sdk/steprun, where 5fbc366f holds the same answer on error,
so G's deletion of buildstep's askrun.go and its test is kept. The tree is G's.
Builds: design 47 G A-286
Claude-Session: https://claude.ai/code/session_01NcbgfGKaxw984XyAPfLm4h
KEPT gains failed-steps, idle and transfers, so each declared state can be read by
its tools; the force-push refusal's test holds that the refusal names the one tool
that may do it and that it runs on the operator's Approve of that call.
Builds: design 47 G A-287, A-288
Claude-Session: https://claude.ai/code/session_01NcbgfGKaxw984XyAPfLm4h
Issue 1409. A stat through root of a path that is not there answered that the file modes refuse it; it says
it does not exist, so design 39 parts §15's "gone" can be read.
Builds: design 39 parts D3 (§15), bullet 3
Issue 1409. A list of a place the mesh keeps a secret, or of a machine's credential places, at any depth,
answers each entry's name, type, owner, group, mode, size and a link's target held to the content rule,
never content, and says secret-place; a process's environ, mem, fd, map_files, root and cwd stay refused.
Lines touched cite ADR 0394 §7.
Builds: design 39 parts D3 (§15), bullets 2 and 5
The seat mesh-pipeline-runner's step event carries `fingerprint`; gitea read
`fp`, which no event carries, and which the type-check refuses once the Node
fetch places the runner's contracts (design 47 G A-285).
Builds: design 47 G A-285
Claude-Session: https://claude.ai/code/session_01NcbgfGKaxw984XyAPfLm4h
gitea imports @novox/mesh-delivery-contracts-mesh-pipeline-runner, so its
package.json names it and the Node fetch places the group's (design 47 G A-285);
the catalog view's revision answer is typed with its revision and entries, which
revisionAnswer already refuses loudly when the holder leaves them out.
Builds: design 47 G A-285
Claude-Session: https://claude.ai/code/session_01NcbgfGKaxw984XyAPfLm4h
The fetch takes `group-packages` (a member's repository, commit, ref and the
package's directory there) in place of the single `package` and its tagged form;
it packs each package a module of the tree takes, publishing nothing, and places
it in every module whose package.json names it, the SDK in every module, in place
of the registry's whatever the lock names.
Builds: design 47 G A-284, A-285
Claude-Session: https://claude.ai/code/session_01NcbgfGKaxw984XyAPfLm4h
mesh/check checking fff3b819d383 on f1c04801e477 | novox/mesh-catalog#663 in group fix/checker-preview-after-failed-step
A level that does not compile or generate ended the check step failed with no
result, so the checker could not tell a merged fetch from one that never
answered, and previewed and judged no manifest or composition of the member.
The step now ends failed with its answer all the same: its fetch and the
levels so far beside the error, which the step's end already keeps.
Builds: design 47 G A-286 (G-22-lead-answers-2.md at hq 41dd30fe; issue 1921; principle 3)
Quick: this checker on main; thorough: the same rule in G's checker (join 11, J15).
Claude-Session: https://claude.ai/code/session_01NcbgfGKaxw984XyAPfLm4h
Try 2 of milestone G failed in go test, verifying nats-server against the checksum database: restic's
integration tests load it through testbus, neither its go.mod nor its go.sum named it, and the fetch's
shut-off load ran untagged, so it never met it. The second failure of this kind (after the join's
dropped zero-version requires), so the guard is held once, where every repository's check passes: each
directory of the checked tree is loaded untagged and with the integration level's tags, and its own
go.mod and go.sum must name every module the loads reach (a mesh module required, no sums; every other
required at the version selected, with both go.sum lines). The fetch fails naming the module and the
line. A sibling laid beside is loaded as before, unjudged.
This guard lands with G, so G's own check still runs today's proof; no harm, the gaps it finds in the
seven repositories are fixed in this join.
Builds: design 47 G (join 11, the check's shut-off build of the group's modules)
Claude-Session: https://claude.ai/code/session_01NcbgfGKaxw984XyAPfLm4h
The check builds with GOPROXY=off, so a module its packages load that go.sum does not hash is looked up in the
checksum database and fails (try 2: verifying nats-server v2.11.17). mesh-sdk testbus, which restic's tests import,
brings nats-server and its four requirements; go.mod and go.sum now name them, as go list -mod=readonly -deps -test
against the group's heads asks (both build tags).
Builds: design 47 G (join 11, the check's shut-off build of the group's modules)
Claude-Session: https://claude.ai/code/session_01NcbgfGKaxw984XyAPfLm4h
Area 2 added seat:operator-channel.notify to the manifest's invokes for the call in main.go,
and the test of the manifest's invokes still listed the old ones, so the unit level failed.
Builds: design 47 G GR8
Claude-Session: https://claude.ai/code/session_01NcbgfGKaxw984XyAPfLm4h
The checker's fetch pins only the zero-version requirements a directory's go.mod names
(mesh-build-go workspace.go goModOf), so a module reached only through another, such as
meshhandacts through meshhandactsclient or mesh-delivery-sdk through meshcatalogclient,
was never downloaded and the shut-off load failed (fetch.go:267). The join's merges had
dropped these indirect requirements that the trunk's go.mod files list.
Builds: design 47 G (join 11, the check's fetch of the group's modules)
Claude-Session: https://claude.ai/code/session_01NcbgfGKaxw984XyAPfLm4h
The conflict in the old cmd/claude-licence-manager/tools_test.go resolves to both sides: main
moved it to cmd/manager, and the join's count of binding.changed.<consumer> under its family
goes into the moved TestResyncSaysEachOnce. The whole-module test's runtime refused
binding.changed.<consumer> because the manifest names the family binding.changed: it now
accepts a key that extends a named family by further parts, as A-134 says the event.
Builds: design 47 G A-134
Claude-Session: https://claude.ai/code/session_01NcbgfGKaxw984XyAPfLm4h
mesh/check being carried to cff1005422c1 on 6d000647de2d | from f9dcbc295723
The one bundle that did everything becomes the manager bundle, which runs long and answers the seat's ten verbs and
the three acting tools, and a tools bundle of its own that answers the four reading tools and writes nothing, so they
answer while the manager fails to load or stops. Both import the module's one package for the layouts of the kept
data and the log entries, the key form, the entries' names and the settings with their validity; the settings file
and the account with its directory stay as they are.
Builds: design 39 parts §1, §2, §3, §4, §5, §6, §7, §8
The gate consumes pull-updated and invokes open-pulls and pull; it writes deliveries and groups
under their key forms, deletes them as a take-in or a closed pull request leaves them, copies a
delivery's base onto the record's member, answers groups from records and the group key together,
puts a missed event right at its start, and writes switch.done once the switch has run.
Builds: design 47 G §716, A-279, A-236, A-281
Claude-Session: https://claude.ai/code/session_01NcbgfGKaxw984XyAPfLm4h
The carry-over from the store database ran live once and carried every row, so its code, the gate that answered
"the carry-over has not ended", the condition carry-over-failed, the tool licences_carried, the opening of grants
with grant.key and the postgres driver leave the module, with its requirement of the database and of the secret
grant-key. The state carried stays as the carry-over's record; at its start the manager waits for nothing and
clears a carry-over-failed it finds kept.
Builds: design 39 parts §6 (the template files/database.url), §9, §10, §11
nats and distribution declare the old stream form and require no finding-events, so their writers stay on it.
Builds: design 47 G A-274 (answers G-22)
Claude-Session: https://claude.ai/code/session_01NcbgfGKaxw984XyAPfLm4h
buildstep no longer takes mesh-sdk's buscalls objects (that package now reaches
the bus): the module judges each object of its kept-builds store with
KeptBuildOf and orders them with SortKept.
Builds: design 47 G (join 11 area 4: buildstep kept free of the bus)
Claude-Session: https://claude.ai/code/session_01NcbgfGKaxw984XyAPfLm4h
AnswerSeat checks a call's caller against the seat's definition its caller
reads from the seat list it follows (A-186). The bus module's keeper hands
it the mesh-broker definition its seat list holds once heard; mesh-recovery,
which followed none, follows one over its connection and an event hub, read
again once its events are live, and its manifest and its process invoke
seat-definitions and consume the list's two events. Its whole test answers
the seat list.
Builds: design 47 G lead answers A-179, A-186 (join 11 area 6)
Claude-Session: https://claude.ai/code/session_01NcbgfGKaxw984XyAPfLm4h
The vendor retired the session limit and answers the session window empty, so Room required a percentage that never
comes: live, no licence had room and no switch could choose a target. Room is now 100 less the weekly percentage; the
session window takes no part in the switch threshold, a target's bound, the near-limit warning or no-room's words.
A reading with no weekly percentage still has unknown room. The settings' descriptions and the README say so.
The whole tests' vendor answers the session window empty, as it does live; on the old rule they failed with the live
symptom ("has no session or weekly window in its reading").
Builds: design 39 bus-and-switch C4
The merge of main's holder test (caller list from the config watch) over C3d's
(no caller list, REG2) left cases the merged holder does not answer; the
holder's code is C3d's, so its test is too, with the moved delivery paths.
Builds: design 47 G REG2 (C3d)
Claude-Session: https://claude.ai/code/session_01NcbgfGKaxw984XyAPfLm4h
The join's importers name the delivery repository's moved contracts and sdk
(MV1), node-tools' tests use mesh-sdk's gate, the duplicate node-engine claim
rule of the reader release gives way to G's, and the assigner names its first
watched branch, never a constant. The seat holders' AnswerSeat callers are
area 6's.
Builds: design 47 G join 11 (merge of fix round 11)
Claude-Session: https://claude.ai/code/session_01NcbgfGKaxw984XyAPfLm4h
Live after #659 the manager, in its own runtime, was still issued none of its logs: the runtime on main issues
template states but logs only from the `logs` section. usage (35 days, 256 MiB) and audit (64 MiB) move there; the
states stay templates. The whole tests' runtime now issues logs only from `logs`, as main does, and failed on the
old manifest.
Builds: design 39 bus-and-switch C3
Each module's puller keeps an answer over what step-ended carries inline in the object store `outputs` its manifest requires
(template objects, 256 MiB) and its tools serve `<seat>.output` from it; the manifest claims the verb.
Builds: design 47 G A-254, A-257
Claude-Session: https://claude.ai/code/session_01NcbgfGKaxw984XyAPfLm4h
Moved to its own account, the manager finds manager-key.json still owned by the tool runner's account, which the
node-engine does not hand over; it skipped making a pair because the file existed, and every verb answered the read
error. Now a pair it cannot read is replaced by a new one, renamed over it, and said in its journal; every use passes
the public key per ask, so nothing depends on the old pair.
Builds: design 39 bus-and-switch C2
Claude-Session: https://claude.ai/code/session_01NcbgfHb9d4T9nq5Q6WqdiJ
Run inside the machine's tool runner, the manager was issued none of its own logs and states, so the carry-over
failed live ("keeps no log called usage ... it was issued: none"). It now declares runs-as claude-licence-manager,
makes that account, owns what it places on its machine and its secrets, reaches the bus with its own broker secret,
and invokes a node's claude_code_grant, which the tool runner's account could ask and its own could not.
The whole tests play the manager's own runtime: they issue only the states, logs, asks and events its manifest
names, and fail on a manifest without runs-as; a unit test holds every ask in the code to the manifest's invokes.
Builds: design 39 bus-and-switch C1
Claude-Session: https://claude.ai/code/session_01NcbgfHb9d4T9nq5Q6WqdiJ
OpenForgeMade is the fill; no path of the gate opens a forge-made record (design 47 G §151 ends the mark from G's delivery),
so the teller says none until one does.
Builds: design 47 G §170, §522, A-161, A-172
Claude-Session: https://claude.ai/code/session_01NcbgfGKaxw984XyAPfLm4h
A catalog revision is pinned by its watched branch and its number (revisions are numbered per branch): PinnedRevision gains
branch, Pin and Free take it, the key names it so two branches' revision 9 are two pins, and pinsOf says {branch, revision}.
Also the C2 merge's sourcetools imports follow the contracts' new path.
Builds: design 47 G §579, A-171
Claude-Session: https://claude.ai/code/session_01NcbgfGKaxw984XyAPfLm4h
The keeper follows meshcatalogclient's branch follower and the seat list on its views reader, and plans again at each
move of a branch's view, the registrations or the seat list. Each local name is planned from the first of the module's
watched branches in its registration's order; another branch asking it with other settings is said drifted, naming
the branch. A module of the new form whose resolved seat definition has a verb or an event gets its sent store; one
including only parts, and a foundation part, none. A module in no watched branch's newest revision loses its plans
key, its entries orphans. The sum of the caps counts the old sections the entries declare. The manifest consumes the
catalog half's and the seat list's events and invokes their verbs; manifest-accepted and indexer_bus_state go.
Builds: design 47 G §344, §433, §532, §533, §537, §538, §547, §591
Claude-Session: https://claude.ai/code/session_01NcbgfGKaxw984XyAPfLm4h
The runtime source is *steprun.SeatAsks (named worker not bound is said through the module's journal); the callers follow
the seat list (steprun.SeatListCallers).
Builds: design 47 G §543, §544, A-163, A-170
Claude-Session: https://claude.ai/code/session_01NcbgfGKaxw984XyAPfLm4h
The config `callers`, the tool registry_callers and its invokes, the caller-list code and the holder's per-verb
Allowed checks go. The windows' driver counts as build modules the modules claiming a `node-build-*` seat in the
catalog half, followed by the seat mesh-catalog's branch follower on the program's views reader (one shared events
hub); it wakes again when a build module's entry moves. The retention check of remove finds the artifact sweep's
holder through the seat mesh-module-assigner's holders (seatlist.HolderOf), not a name.
Builds: design 47 G §610, §7, §428, A-156, A-82
Claude-Session: https://claude.ai/code/session_01NcbgfGKaxw984XyAPfLm4h
An integration test starts the night as the node does: its credential is read from where the manifest places it (the process
night's MESH_BROKER_FILE, the module's own-secrets broker), it says each item started and ended on the seat node-backup, gives
way to a restore that runs (items put off as keys of the state put-off, started again on the restored) and, past put-off-bound,
the bus's expiry writes backup-waits into the findings. It found that WaitRestored held a fetch of a minute after its context
ended, so an expiry ended the wait up to a minute late: it now consumes the restored events and ends on the context.
Builds: design 47 G §563, §561, §554, A-150, A-155
Claude-Session: https://claude.ai/code/session_01NcbgfGKaxw984XyAPfLm4h
The record backed-up reads is the key `<module>.<item>` of the module's own state put-off (key-ttl), bounded by the setting
put-off-bound (minutes, default 60); the bus removing it is heard as the expiry, which writes the finding backup-waits through
the held writer over the module's findings and ends the wait, so the night never holds its lock without a bound. The file
put-off.json and its writes are gone. The manifest requires put-off, findings and finding-events and hears findings-asked.
Builds: design 47 G §561, §563, §557, §302, A-155, A-77
Claude-Session: https://claude.ai/code/session_01NcbgfGKaxw984XyAPfLm4h
All of mesh-build-go's buildstep Go module leaves mesh-catalog for
modules/mesh-delivery-sdk: the step run, fetch and release and the
regenerate-and-compare step as the package steprun, the CPU share as
steprun/buildcpu, the conformance kit as steprun/conformance, the seat
generator as seatgen with cmd/seatgen, and cmd/mesh-pack. The three
build modules and the builder's and checker's tests import the SDK and
keep no copy; the SDK requires nothing of mesh-catalog. Code the build
modules share is the delivery domain's SDK, not one module's package.
It carries the build-module round's two new files of that package
(dropdetected.go 247c422f, storedbuilds.go 904b8aed).
Builds: design 47 G §464, §680, §327, §420, §495
Claude-Session: https://claude.ai/code/session_01NcbgfGKaxw984XyAPfLm4h
The tool is marked in the manifest's needs-approval, so each call runs on the operator's Approve; the manifest
uses operator-channel; set_full keeps refusing a force-push onto a trunk and names the new tool; both clients
gain deleteBranchProtection.
Builds: design 46 approval mark §21 (milestone G)
Claude-Session: https://claude.ai/code/session_01NcbgfGKaxw984XyAPfLm4h
A Go program is linked with -X main.meshVersion beside main.builtFor, a Node program's package.json version is set, an
image is labelled org.opencontainers.image.version; a build with no version stamps none.
Builds: design 47 G §458, §573
Claude-Session: https://claude.ai/code/session_01NcbgfGKaxw984XyAPfLm4h
The guard's toolchains value carries the sha256 of the go or node program found on the path (read again when the version
changes); toolchain-reported is said on the build seat's subject of the node at start and on each change of version,
recipe or digest; a build's ToolchainUsed names the digest. The legacy toolchain.reported emit and its stream go: nobody
consumes them.
Builds: design 47 G §492, §472, §466 (lead answer A-152)
Claude-Session: https://claude.ai/code/session_01NcbgfGKaxw984XyAPfLm4h
The delivery repository's contracts and SDK live under modules/ as
mesh-delivery-contracts and mesh-delivery-sdk, their Go module paths
renamed to match the folders (the TypeScript names already did), and
every importer in mesh-delivery, mesh-catalog, mesh-host, mesh-tools
and mesh-sdk follows in the same group: a design as it should be, one
path per module, checked by the group's check building every importer.
Builds: design 47 G §427, §88
Claude-Session: https://claude.ai/code/session_01NcbgfGKaxw984XyAPfLm4h
Each member's build is read at /v2/<module>/<artifact>/blobs/<digest> through the required provision artifact-store
(bound to a file), checked against its digest and laid as the package it names: a Go module zip replaced in the temporary
modfile, an npm tarball in node_modules. A mismatch or a store that does not answer fails the step naming the member.
Builds: design 47 G §465, §373, §466 (lead answer A-151)
Claude-Session: https://claude.ai/code/session_01NcbgfGKaxw984XyAPfLm4h
A drop of a kept detected-steps key records the act with the caller the bus proved, then deletes the key; a log that
refuses deletes nothing and is answered error; a drop without why is refused; fetch is asked unless drop.
Builds: design 47 G §463, §480
Claude-Session: https://claude.ai/code/session_01NcbgfGKaxw984XyAPfLm4h
mesh/check being carried to 15db34169e24 on 89ad37048520 | from d9f2b7bcfa07, mesh-contracts 29ec1c3df312
mesh-delivery's parts.Receiver.Take took a chunk 0 again on purpose (276412d7, design 51 Publishing, 1966/AC203-1,
AC203-2); the holder's test still expected it refused and failed on every run, on trunk too. The test now expects a
chunk out of sequence refused and a chunk 0 again taken.
Builds: design 51 Publishing, the bullet "The file travels in parts through the seat"; issue 1975's group check
Claude-Session: https://claude.ai/code/session_01NcbgfGKaxw984XyAPfLm4h
The whole test runs the process on its own account against a bus of its
own, every bound a setting of seconds, with a stand-in bus module for
snapshot-get, snapshot-restore and snapshot-restored: a hand take's log,
a restore waiting on the dump past its bound key, busy and the ask again,
the expiry on its bound key, a keeper's snapshot in three chunks, a
protected restore cancelled at its bound and its aside put back, one the
keeper answers neither, and a stop putting the aside back.
It found that an act said its ending event while it was still the running
one, so a caller told busy that asked again on hearing it was told busy
with the same act and heard nothing more: the act is now released and
written ended before its ending event, and the removal at a snapshot-kept
runs before its ended.
Builds: design 47 G §557, §554, §681, §562, §551, A-138, A-140, A-141, A-143
Claude-Session: https://claude.ai/code/session_01NcbgfGKaxw984XyAPfLm4h
mesh/delivery-group group fix/1975-kernel-log rejected: composed together: the checker's verdict at the members' heads (c265ee221ccf): unit: distribution/inter…
mesh/check failed (2): unit: distribution/internal/holder.TestTheHolderOfTheTwoImageSeats/a_transfer_in_par…; and 1 more | ebcc7761f178 on d9f2b7bcfa07
Each wait's bound is a key of the module's state bounds with its own time to
live, dump-wait.<act> and restore-wait.<act>, its expiry heard on the key's
watch, so no timer stays in the process.
Builds: design 47 G A-143, A-77, A-138, §681, §561
Claude-Session: https://claude.ai/code/session_01NcbgfGKaxw984XyAPfLm4h
point-keep-days, dump-wait-minutes and snapshot-restore-wait-minutes may
be fractions, and expire-every-hours (24 unless set) times the removal of
old points by a key of the state bounds whose own time to live the bus
ends, heard as its change: no timer in the process.
Builds: design 47 G §551, §557, A-77, A-130, A-138
Claude-Session: https://claude.ai/code/session_01NcbgfGKaxw984XyAPfLm4h
journalctl --dmesg --boot=N under the same escalation, window, priority, fixed-string match, 2000-line cap
and redaction as the journal verb (shapeLines is the one filter both use); a boot is 0 or a negative whole
number, refused otherwise before anything runs. No scope: the kernel belongs to no manager.
Builds: issue 1975, the operator's request of 2026-11-12 (the lead's handover; no design line names the verb)
Claude-Session: https://claude.ai/code/session_01NcbgfGKaxw984XyAPfLm4h
A move or a restart stops the process: from its stop signal every call
is answered an error naming the stop, the running act ends through its
own failure path (an aside put back whole, whatever its context), and
every accepted act not ended ends failed with stopped and its event, so
its caller asks again.
Builds: design 47 G §562, A-141
Claude-Session: https://claude.ai/code/session_01NcbgfGKaxw984XyAPfLm4h
An unprotected snapshot is received through snapshot-get into
incoming/<request>, refused before the first chunk past the cap, checked
against its digest before any aside, and restored by the module; a
protected one is verified, the running data set aside and restored in
place by the keeper, waited on by snapshot-restored within
snapshot-restore-wait-minutes, cancelled at the bound. The bus module
sends its chunks with mesh-sdk's form.
Builds: design 47 G §681, §682, A-137, A-138, A-139, A-140
Claude-Session: https://claude.ai/code/session_01NcbgfGKaxw984XyAPfLm4h
The validator refused it as placed nowhere (ADR 0348 §5): the keeper reads it from its settings
beside ask-bound-seconds and view-waits-seconds, which are marked so.
Builds: design 47 G §557
Claude-Session: https://claude.ai/code/session_01NcbgfGKaxw984XyAPfLm4h
The manager says binding.changed.<consumer>: the module's views reader hears it by the adoption
and asks the first view (FirstBinding) once its subscription is taken, instead of a subscription
of the exact subject of its own.
Builds: design 47 G A-134
Claude-Session: https://claude.ai/code/session_01NcbgfGKaxw984XyAPfLm4h
The follower main added (issue 1409) used the removed views.FindingWriter: it is adopted by a
views.Reader holding the vault's one finding writer, which the assignments view shares, and hears
claude-licence-manager.binding.changed.<consumer> through the adoption.
Builds: design 47 G A-16, A-28, A-134
Claude-Session: https://claude.ai/code/session_01NcbgfGKaxw984XyAPfLm4h
The consumer is the event's further part, the one form of G's seat events, so mesh-vault and
claude-code adopt it through their views reader.
Builds: design 47 G A-134
Claude-Session: https://claude.ai/code/session_01NcbgfGKaxw984XyAPfLm4h
Its tool check stands on its own runtime beside the process's unit check, its
template states are its own by requires alone (not listed again in the
process's bus, which keeps its account), and digests takes the state
template's fixed cap (A-135).
Builds: design 47 G §657, §567, §554, A-133, A-135
Claude-Session: https://claude.ai/code/session_01NcbgfGKaxw984XyAPfLm4h
claude-code's and mesh-vault's manifests take both sides' additions (G's seat reads and
findings-asked, main's licence seat verbs and events); claude-code's manifest test lists both.
mesh-vault's licence-bindings follower still uses the removed views.FindingWriter: fixed after.
Builds: design 47 G §610
Claude-Session: https://claude.ai/code/session_01NcbgfGKaxw984XyAPfLm4h
The catalogue keeps no file text inline for a new module: each setting's file is a template
under files/, packed by the module's archive artifact. A local build of the tools bundle
leaves a binary the module's ignore file now names.
Builds: design 47 G §633, A-112, A-130
Claude-Session: https://claude.ai/code/session_01NcbgfGKaxw984XyAPfLm4h
The module measures its points directory against the fill mark of its cap at its start and
after each act, writes recovery-waits for an act that waits on the dump past its bound and
clears it once the act starts its work or ends, and writes aside-unrestored for each aside
only it holds the data of, cleared once the aside is restored or discarded. The bounds are
the settings points-filling-mark and dump-wait-minutes, read at the start.
Builds: design 47 G §561, §557, §564, §633, A-112, A-130
Claude-Session: https://claude.ai/code/session_01NcbgfGKaxw984XyAPfLm4h
The process reads its broker account, answers the seat mesh-recovery's verbs itself with the
caller the bus proved, takes and restores streams through mesh-sdk's bussnap, keeps the digests
in its state, names a kept data's bus objects from its catalog entry as mesh-sdk names them,
learns the fingerprint a module runs at from the node-engine seat's running and the bus
module's dumps from the backup holders' item events, each read whole at its start, and asks
the backup holder on its node to bring the points directory back. A restore whose
restore-started the bus did not acknowledge reads and sets aside nothing.
Builds: design 47 G §554, §516, §551, §556, §563, §567, §655, A-107, A-112, A-121, A-123, A-126, A-132
Claude-Session: https://claude.ai/code/session_01NcbgfGKaxw984XyAPfLm4h
The module's declared health check names recovery_kept_points: it answers what kept-points
answers with the health answer beside it, unhealthy while an aside only it holds the data of
is listed.
Builds: design 47 G A-129, §554, §657
Claude-Session: https://claude.ai/code/session_01NcbgfGKaxw984XyAPfLm4h
The take and restore of one stream is shared code in mesh-sdk; the bus module's keeper, its
nightly dump and its tools import it from there and keep no copy.
Builds: design 47 G §655, A-125
Claude-Session: https://claude.ai/code/session_01NcbgfGKaxw984XyAPfLm4h
Builds: design 47 G §554, §656, §657, §682
recovery_kept_points reads kept-points; every hand tool (take, restore, retry, aside, discard_aside, restore_aside, cancel, expire_points) is one verb of the seat called with the asker the tool runner proved; the bundle holds no grant of its own.
Claude-Session: https://claude.ai/code/session_01NcbgfGKaxw984XyAPfLm4h
Builds: design 47 G §516, §554, §656, §682, §691, A-107, A-115
Each acting verb answers one of accepted, busy or error; arguments are read afresh per call and refused when unknown (local-name, from, aside-only, discard-aside) or without why; restore takes exactly one of snapshot and request (request is refused until built); the hand-act log names the proved account and the asker.
Claude-Session: https://claude.ai/code/session_01NcbgfGKaxw984XyAPfLm4h
Builds: design 47 G §516, §554, §566, A-123, A-124
aside takes the bus names from the catalog entry of the fingerprint the node-engine's placement runs, takes the aside whole and checked, removes the live objects, says set-aside; restore-aside puts it back.
Claude-Session: https://claude.ai/code/session_01NcbgfGKaxw984XyAPfLm4h
Builds: design 47 G §516, §554, §560, §562, §563, §564, §566, §567, §682, §691, A-117
restore says restore-started, waits for the dump, finds the point (or has the backup holder bring it back), checks its digests against the bus, sets the live data aside whole and checked, removes it, restores the point and its consumers, and puts the aside back when a step after the removal fails; cancel ends a waiting act failed/cancelled with the hand act's words; at a start it clears temporary points, puts an aside back and ends a cut-off act failed/stopped.
Claude-Session: https://claude.ai/code/session_01NcbgfGKaxw984XyAPfLm4h
Builds: design 47 G §516, §551, §554, §560, §564, §567, §517
Accept writes the hand-act log first, answers accepted/busy/error at once and writes the act before it answers; take writes each stream under a temporary name, its digests on the bus before ended; the cap is refused before anything is written; expiry never removes the newest point.
Claude-Session: https://claude.ai/code/session_01NcbgfGKaxw984XyAPfLm4h
Builds: design 47 G §554, §656, §657, §633, §564, §557, A-107, A-115
The seat claimed once for the mesh with every verb of its definition; a process answering them (serves beside bus), a tools bundle, the data item points, the state digests (§567), the findings of §561, the four settings.
Claude-Session: https://claude.ai/code/session_01NcbgfGKaxw984XyAPfLm4h
The lead's answer B44: a lease the bundle before held in today's database
and still live at the run is written to `leases` under the bus's key form
with what is left of its lifetime, so no refresh races it whatever the
delivery's order; an expired one is counted read and not written.
Builds: design 39 bus-and-switch §4, §12, B44
Claude-Session: https://claude.ai/code/session_01NcbgfHb9d4T9nq5Q6WqdiJ
Review round 1, finding 5: the reading reads the grant under the licence's
lease (another run holding it reads nothing that round), and a grant the
vault refuses writes readError, keeps the percentages and goes on to the
next licence, so the conditions are still said after the round.
Review round 1, finding 6: these answers were built by earlier commits that
did not name them: B26, B27, B28 and B29 by M-6 (9cc62818), B30 by M-9
(39912119), B35 by M-11 (6cf98386); B23 and B24 by V-1 (e85cf002) and M-7
(5ca3876b), named again in b8e8be61; B31 and B32 in 36d9d739.
Builds: design 39 bus-and-switch §4, §16, §24, §31, B23, B24, B26, B27, B28, B29, B30, B34, B35
Claude-Session: https://claude.ai/code/session_01NcbgfHb9d4T9nq5Q6WqdiJ
Review round 1, finding 4: the vault hands a licence's access token only to
a machine a binding names, so licences_resync builds it from the grant (the
key itself for an API key) instead of reading it back.
Builds: design 39 bus-and-switch §15, §17, §42, B23, B24
Claude-Session: https://claude.ai/code/session_01NcbgfHb9d4T9nq5Q6WqdiJ
Review round 1, finding 3: adoptedFrom is the node a login came from, or
where an API key came from; a newer login of the same account keeps the
latest reading and generation the key holds.
Builds: design 39 bus-and-switch §2, §34, §46
Claude-Session: https://claude.ai/code/session_01NcbgfHb9d4T9nq5Q6WqdiJ
Review round 1, finding 2: a reader's first view taken while the carry-over
runs is brought up to date by the event said after each key is written.
Builds: design 39 bus-and-switch §12, §21
Claude-Session: https://claude.ai/code/session_01NcbgfHb9d4T9nq5Q6WqdiJ
Review round 1, finding 1, and the lead's answer B43. Until the vault's first
view of the licence bindings is answered since its start, licence-put's
access token is refused, saying why. A first view removes a consumer's key it
does not name only when that key was written before the view was asked. The
newer generation's binding staying over a first view (B32) was built in V-2's
keep and is named here.
Builds: design 39 bus-and-switch §17, B31, B32, B43
Claude-Session: https://claude.ai/code/session_01NcbgfHb9d4T9nq5Q6WqdiJ
vault_first_views lists the licence bindings' first view beside the
assignments'; vault_first_view_again asks the one its new argument `of`
names, and both when it is absent.
Builds: design 39 bus-and-switch §52, §15 to §17, B21, B42
Claude-Session: https://claude.ai/code/session_01NcbgfHb9d4T9nq5Q6WqdiJ
The built binary played as its runtime plays it, its kept data in a real
nats-server's buckets and streams, today's rows in a real postgres: a start
stopped at a table says carry-over-failed and acts on nothing, the next
carries every row and a further one writes nothing; an unreachable database
calls no vendor; a reading at the threshold moves four consumers in one act
and no-room is raised and cleared on the operator channel's subject, read
through its client; a new login is adopted, listed and a target.
Builds: design 39 bus-and-switch §9, §11, §12, §13, §23, §26, §28, §29, §46, B41
Claude-Session: https://claude.ai/code/session_01NcbgfHb9d4T9nq5Q6WqdiJ
The licences verb answers each licence's organisation, latest usage with its
reset times, its room or why it is not known, whether it is where a full
licence's consumers move next, when and from where it was adopted and its open
conditions. The usage verb takes since, a limit of at most 2000 and raw. The
module's tools: licences_switches from the audit, licences_read_usage and
licences_switch_now with dry, licences_settings with the level mesh-config
answers (the manifest invokes its config verb), licences_kept over every
bus-state entry with its fill, licences_resync, licences_conditions and
licences_carried with its run, refused once carried. An acting tool waits for
the carry-over.
Builds: design 39 bus-and-switch §34, §35, §37, §38, §39, §40, §41, §42, §43, §44, §47, §20, §11, B9, B18, B40
Claude-Session: https://claude.ai/code/session_01NcbgfHb9d4T9nq5Q6WqdiJ
Each condition the manager holds is written in its state conditions with its
body and since, then said on the operator channel's condition accept under
claude-licence-manager, scope module; it is said again only when its words
change, and said cleared, its key deleted, once it no longer holds. At its
start, after the carry-over's run, it judges every kind anew: raised again
each it holds, cleared each kept key it no longer holds. The kinds:
carry-over-failed, settings.<setting>.invalid for a pair above
switch_at_pct and for a setting outside its bound, and per licence no-room
when the rule finds no target, near-limit at warn_at_pct with consumers, and
usage-unread once its latest success, or its adoption when none, is older
than usage_stale_minutes. Headlines name counts and short names, cut at 60.
Builds: design 39 bus-and-switch §9, §11, §29, §30, §31, §33, B17, B34, B36, B37, B38, B39
Claude-Session: https://claude.ai/code/session_01NcbgfHb9d4T9nq5Q6WqdiJ
After each reading of a licence with consumers whose session or weekly
window is at or above switch_at_pct, the manager moves every consumer bound
to it, when switch_auto is on and the settings are valid, to the other
licence with known room whose windows are both below room_below_pct and the
most room, the first by name on a tie. Each consumer moves as the switch verb
does, by usage, and says binding.changed; the act says one licence.switched
and one audit entry switched with the window, its percentage and reset and
the target's room, and is told to the operator; the switch verb is not told.
By hand the rule runs with switch_auto off, and dry answers the moves alone.
Builds: design 39 bus-and-switch §26, §27, §28, §32, §33, B22
Claude-Session: https://claude.ai/code/session_01NcbgfHb9d4T9nq5Q6WqdiJ
The settings gain switch_auto, switch_at_pct, room_below_pct, warn_at_pct,
usage_minutes and usage_stale_minutes with the manifest's defaults and
bounds, a value outside its bound keeping the default. The settings file is
read again at every reading, each changed value written in the audit under
settings, and the usage timer follows usage_minutes. A room or warning above
switch_at_pct is found as an invalid pair naming both values.
Builds: design 39 bus-and-switch §33, §24
Claude-Session: https://claude.ai/code/session_01NcbgfHb9d4T9nq5Q6WqdiJ
grantedBy is dropped: any caller of snapshot-get gets the chunks, a protected snapshot never.
The chunks are a transfer on the core bus, published with a core publish and kept on no stream.
Builds: design 47 G A-109 (G-20-lead-answers), §681
Claude-Session: https://claude.ai/code/session_01NcbgfGKaxw984XyAPfLm4h
snapshot-get, snapshot-restore, snapshot-restore-cancel and snapshot-remove are answered by
the keeper's process with the proved caller; it publishes snapshot-get's chunks and hears
their acks on its own subscription, and acts at once with no hand-over through requests
(the record of a running restore is written before it runs). The tools bundle keeps its
tools, nats_snapshots among them. The protected refusal and the whole test's case (send,
restore in place, removal) stay.
Builds: design 47 G A-107 (G-20-lead-answers), §681
Claude-Session: https://claude.ai/code/session_01NcbgfGKaxw984XyAPfLm4h
The usage reading flattens the session, weekly and sonnet windows each with
its reset time, keeps it beside the licence with when it was read, and says
it as usage.read; a failed answer keeps the last percentages and writes the
vendor's status and reason. A licence's room is 100 less the higher of its
session and weekly windows, known only for a subscription read within
usage_stale_minutes whose refresh is healthy; an API key's is never known.
Builds: design 39 bus-and-switch §24, §25, §48, §2
Claude-Session: https://claude.ai/code/session_01NcbgfHb9d4T9nq5Q6WqdiJ
The manager consumes the agent module's holdings-changed, keeping each report
as its node's latest and considering them at once, in place of reading the
agent module's state; once it hears them, at every start after its
carry-over, it says holdings.asked with its time and who asks, so every node
says what it holds again.
Builds: design 39 bus-and-switch §23, §46, B12, B22
Claude-Session: https://claude.ai/code/session_01NcbgfHb9d4T9nq5Q6WqdiJ
The agent module asks the seat's current which licence it holds and its key,
then the vault's get for that licence's access secret sealed to its own key:
at its start, on a binding.changed with a newer generation than it applied,
and on the vault's secret-changed for the licence it holds. The applied
binding keeps the licence's key, so the vault's event is matched by the
carried key, never one formed here.
Builds: design 39 bus-and-switch §19, §18, B7, B9, B11
Claude-Session: https://claude.ai/code/session_01NcbgfHb9d4T9nq5Q6WqdiJ
The agent module consumes the manager's binding.changed, acting only on the
one naming its node's consumer, and each time it hears those events anew asks
the seat's bindings for its own consumer as its first view; its read of the
manager's bindings state goes.
Builds: design 39 bus-and-switch §22, B9, B10, B22
Claude-Session: https://claude.ai/code/session_01NcbgfHb9d4T9nq5Q6WqdiJ
The vault's get judges a licence's secret by a rule of its own, before the
asker rule: a grant goes to any proven caller, and an access token to a
machine on which a consumer's binding in licence-bindings names that
licence's key; a refusal names the rule and is recorded.
Builds: design 39 bus-and-switch §16, §17, A1, B3, B8, B25
Claude-Session: https://claude.ai/code/session_01NcbgfHb9d4T9nq5Q6WqdiJ
The vault consumes the licence manager's binding.changed and, at its start,
asks the seat's bindings until it answers; it keeps each consumer's licence,
the licence's key and the generation in its own state licence-bindings,
never an older generation, deleting a released consumer's key. The state is
in the vault's old state section, and holds no secret.
Builds: design 39 bus-and-switch §17, B2, B9, B10, B22, B25
Claude-Session: https://claude.ai/code/session_01NcbgfHb9d4T9nq5Q6WqdiJ
Each binding the verb bindings answers carries its licence's key beside the
§3 fields, as binding.changed does, so the vault's first view matches a
token's secret name to the binding without the key form of its own.
Builds: design 39 bus-and-switch §17, B10, B25 (hq 128236e7 and after, PR #561)
Claude-Session: https://claude.ai/code/session_01NcbgfHb9d4T9nq5Q6WqdiJ
Bind, switch, release, a login's move and a rotation's advance each say one
binding.changed per consumer moved, with the licence's key, after the key of
bindings is written or deleted; a switch through the verbs says one
licence.switched with the from, to, consumers, by and why the audit's
switched entry also holds. licence.bound, released and rotated are said no
more. An adoption advances the consumers already bound before it moves the
login's nodes, so each is said once.
Builds: design 39 bus-and-switch §21, §27 (the verb's switch), §36, B7, B16
Claude-Session: https://claude.ai/code/session_01NcbgfHb9d4T9nq5Q6WqdiJ
At its first start the manager reads the six tables of its database, read
only, and writes them into its kept data and its grants into the vault at
generation 1, then carried.postgres; until then only its read verbs answer.
A second start never opens the database, and a stopped run is done again
whole without doubling a log entry. The store's write code and its postgres
test go: nothing writes the database again.
Builds: design 39 bus-and-switch §1, §8, §11, §12, §13, A3, B14
Claude-Session: https://claude.ai/code/session_01NcbgfHb9d4T9nq5Q6WqdiJ
Recovered from cfee5f2d (feat/47-G-buildmods).
Builds: design 47 G §571 (the build writes a mesh package's lock entry from the latest published version), A-56
Claude-Session: https://claude.ai/code/session_01NcbgfGKaxw984XyAPfLm4h
Recovered from 90d70b83 (feat/47-G-asks).
Builds: design 47 G §40; design 46 §6 Limits, batching, history (no ask hidden or refused for its asker's count)
Claude-Session: https://claude.ai/code/session_01NcbgfGKaxw984XyAPfLm4h
Every reading is appended under its licence's key with the vendor's answer whole (ADR
0054), and every act under its kind, never rewritten; an act whose audit entry the log
does not take fails and says so. The memory store goes: the tests run the bus store over
a bus kept in memory.
Builds: design 39 bus-and-switch §6, §7, B20
Claude-Session: https://claude.ai/code/session_01NcbgfHb9d4T9nq5Q6WqdiJ
A lease is a key created with a two-minute lifetime, which fails while another holds it:
the bus's own compare-and-set keeps one rotation source (ADR 0183 §6). An offered grant's
outcome is kept under its fingerprint in the key form for 90 days, every outcome kept.
Builds: design 39 bus-and-switch §4, §5, B13
Claude-Session: https://claude.ai/code/session_01NcbgfHb9d4T9nq5Q6WqdiJ
The state licences holds one key per licence by its key form, never a secret, with its
latest reading beside it; the state bindings, the same local name the consumers read,
holds each consumer's licence, kind, generation, since, why and by, written by the store
itself. Each consumer counts its own generations, past what its node reported.
Builds: design 39 bus-and-switch §2, §3, B11, B22
Claude-Session: https://claude.ai/code/session_01NcbgfHb9d4T9nq5Q6WqdiJ
The manager hands what the vendor answers to licence-put, the grant first and then the
access, at the licence's next generation, and reads a grant with get for one act; it keeps
no token at rest (ADR 0550 §3). The sealing at rest goes; opening stays for the carry-over.
Builds: design 39 bus-and-switch §15, §16, §20, §45, B15
Claude-Session: https://claude.ai/code/session_01NcbgfHb9d4T9nq5Q6WqdiJ
The vault hands every token (ADR 0550 §3); current names the licence, its key, kind,
generation, health and access expiry, and keeps the account's identity for the agent's
account file. bindings takes an optional consumer, answered unpaged.
Builds: design 39 bus-and-switch §18, B7, B10, B19
Claude-Session: https://claude.ai/code/session_01NcbgfHb9d4T9nq5Q6WqdiJ
The licence manager follows the nodes' holdings by an event and asks for them with
holdings.asked, instead of reading the agent module's state (ADR 0596 §2). The state
holdings stays the module's own.
Builds: design 39 bus-and-switch §23
Claude-Session: https://claude.ai/code/session_01NcbgfHb9d4T9nq5Q6WqdiJ
A licence's grant and access token move into the vault (ADR 0550 §3), handed in two steps
with a key made for the call, as move-in does; the caller's generation is the record's,
an older one is refused, and secret-changed names the secret and its generation.
Builds: design 39 bus-and-switch §15, B3, B4, B5, B6
Claude-Session: https://claude.ai/code/session_01NcbgfHb9d4T9nq5Q6WqdiJ
Pull request facts come from seat git's pull, ancestry from contains, the review time from the runner's review step,
and get shows one pipeline row per module (no run where the runner's list answers none).
Builds: design 47 G §654, A-75
Claude-Session: https://claude.ai/code/session_01NcbgfGKaxw984XyAPfLm4h
Eight bus-state entries in the typed long form; the old state and the reads of
claude-code.holdings go, and so does the run-once prepare, since nothing writes the
tables again. The database and grant-key stay as the carry-over's read-only source.
The manifest declares its new events, its consumption of holdings-changed, the
operator channel it uses, the verbs it invokes, six settings and its eight tools.
Builds: design 39 bus-and-switch §1, §9, §21, §23, §27, §33, §37-§44, A3, B9, B16
Claude-Session: https://claude.ai/code/session_01NcbgfHb9d4T9nq5Q6WqdiJ
sudo and localization expect version 0.1; distribution and gitea read the old objects and state sections; mesh-build-go, mesh-knowledge, mesh-provisioner and mesh-self-check expect the findings asked in requires and the keeper's ask consumed. Each assertion failed against the changed manifest before it was changed.
Builds: design 47 G §368, §529, §584, A-30
Claude-Session: https://claude.ai/code/session_01NcbgfGKaxw984XyAPfLm4h
A key of the bus takes fewer characters than a licence's name; the key is the store's
alone, and two names that write the same key would overwrite each other.
Builds: design 39 bus-and-switch §10, B6
Claude-Session: https://claude.ai/code/session_01NcbgfHb9d4T9nq5Q6WqdiJ
The healers' manifest already asks findings (template state) and finding-events (template events, the three finding events,
7 days, 16 MiB) and declares no old stream or findings state, and their writer reaches the state by its local name (built in
76be864b and the join); this adds the test that holds it.
Builds: design 47 G §529, §577, A-30
Claude-Session: https://claude.ai/code/session_01NcbgfGKaxw984XyAPfLm4h
The seat mesh-healers' attempt-changed carries the attempt whole with until (a week from its write), on the subject keyed by
the condition and the attempt, each escaped as a key's name is. The event is the seat definition's; the manifest's emits stays
empty as for healer-acted.
Builds: design 47 G §541, §665
Claude-Session: https://claude.ai/code/session_01NcbgfGKaxw984XyAPfLm4h
The node-engine's composed, running, wait-held and wait-ended make the have the hold compares with the wants; the healers'
attempt-changed makes an attempt, dropped at its until on the keeper's own clock (a bound with its own time to live) or when
removed. The manifest's reads are gone and its consumes gain the five events.
Builds: design 47 G §541, §665, §556, §499, §434
Claude-Session: https://claude.ai/code/session_01NcbgfGKaxw984XyAPfLm4h
A findings-stated carrying kinds sets the keeper's view of that observer, node and set of kinds, and clears only the
conditions of those kinds it does not name; no kinds covers every kind. The finding events are taken from either form of the
part's finding stream, the observer and node read from the subject.
Builds: design 47 G §580, §529 (A-30), §383, §384
Claude-Session: https://claude.ai/code/session_01NcbgfGKaxw984XyAPfLm4h
The keeper copies owner and waiting-on from the finding at raising (and into a condition kept from before G at its next
change), hands a pipeline-failed to operator-channel after the setting owner-handover-seconds (default two hours, timed
from the finding's since as a bound with its own time to live), says each change as the change owner, and the tool
conditions_owner sets an owner now, written in the hand-act log.
Builds: design 47 G §540, §685, §678, §557, A-58 (seed-waits: the indexer sets the finding's waiting-on; no kind rule here)
Claude-Session: https://claude.ai/code/session_01NcbgfGKaxw984XyAPfLm4h
A module naming only the mesh's packages beyond typescript and the types needs no lock (gitea names the contracts' git and mesh-catalog packages); a row in the first test fails without the change, as does the catalogue's own run.
Builds: design 47 G A-56, ADR 0469 §5
Claude-Session: https://claude.ai/code/session_01NcbgfGKaxw984XyAPfLm4h
incoming goes to the old objects section; transfers, idle (key-ttl), holds, windows, mirrors and findings to the old state section; the stream events becomes the old stream distribution-events (A-54, EVENTS being the mesh's own stream).
Builds: design 47 G §584, A-54, ADR 0619 §2
Claude-Session: https://claude.ai/code/session_01NcbgfGKaxw984XyAPfLm4h
FindingScope is delivery, the scope the mesh already uses for the delivery's findings; the test names the literal (it failed with gate).
Builds: design 47 G §592, A-42
Claude-Session: https://claude.ai/code/session_01NcbgfGKaxw984XyAPfLm4h
ask-bound-seconds and view-waits-seconds are read by the keeper's own code at its start (settings.go ReadSettings, stdio.Config), so the manifest marks them runtime: true; go.mod requires meshcatalogclient at the zero version, which cmd/mesh-conditions/branches.go imports (the requiredmodules check).
Builds: design 47 G §557
Claude-Session: https://claude.ai/code/session_01NcbgfGKaxw984XyAPfLm4h
transfers, idle (key-ttl), failed-steps (history 1) go from requires entries to the old state section and incoming to the old objects section, as §584 has the forge module keep every state in its old sections.
Builds: design 47 G §584, ADR 0619 §1
Claude-Session: https://claude.ai/code/session_01NcbgfGKaxw984XyAPfLm4h
claude-code, mesh-build-go, mesh-provisioner and mesh-self-check drop their old state findings and stream <module>-findings for the requires entries findings (state) and finding-events (events: the three finding events, 7 days, 16 MiB); mesh-knowledge gains finding-events and the consume. The manifest forms follow mesh-watcher's. The judge refused each for the old form (§529, §577) or for the missing stream or consume (§387).
Builds: design 47 G §387, §529, §577, A-30
Claude-Session: https://claude.ai/code/session_01NcbgfGKaxw984XyAPfLm4h
mesh-delivery-builder.builder_versions answered no numbered version for any name asked (mesh-sdk go and ts, mesh-host, node-tools, mesh-delivery-builder, mesh-conditions), so every module takes 0.1, the floor §368 gives a module with none published: all 144 manifests (141 from "1", distribution, mesh-provisioner and mesh-provisioner-client from "2", which carries nothing over, §583).
Builds: design 47 G §368, §411, §583, A-46
Claude-Session: https://claude.ai/code/session_01NcbgfGKaxw984XyAPfLm4h
Adopt subscribes a seat's event with further parts (<seat>.<event>.>), which never matches the plain views event of an owner; the reader's Follow hears it and asks again on a gap. Also the distribution whole test counts its own two events by subject, as the writer's are said beside them.
Builds: design 47 G A-28, A-26, A-30
The held writer reads the keys of its state when it is made, so it is made after Serve answers state calls; the keeper says a finding is not said before it exists.
Builds: design 47 G A-32, A-29, A-37
The test bus made buckets from the old state section alone, so a template state (deliveries, groups, owed, conditions) was 'not declared'; it now
makes a bucket for each requires entry of the template state under its local name (A-20).
Builds: design 47 G §613, A-20, A-33, A-34 (G-17, G-20)
Removes holder, table, approval, ready, onestatus, pullkey, checks, group_merge, the effects, the forge reads, the teller's
and the view's reads, kvOperator, kvPulls, delivery_merge_again and their tests. deliveries, show, groups and stalled answer
the records as kept (stalled: a record open past a day, and an act owed unanswered past 15 minutes, read from the record's
acts); the switch reads the old section deliveries and says approval-carried for each unperformed merges key. The manifest
drops pulls, asks, reads, the forge, teller and view consumes and invokes, and the events group and merged.
Builds: design 47 G §172, §431, §592, §593, §613, A-24, A-35 (G-3, G-9, G-16, G-17, G-20)
Claude-Session: https://claude.ai/code/session_01JG73KEq4zoDfM9WtYpcQgZ
The writer listens to mesh-conditions.findings-asked, so each also consumes it, as the modules that moved before.
Builds: design 47 G A-30, §383, §686
Claude-Session: https://claude.ai/code/session_01JG73KEq4zoDfM9WtYpcQgZ
WriterConfig.Asked is the generated mesh-conditions contract's SubjectFindingsAsked (A-17), a reader's Findings is the
module's writer itself (A-22, A-26), and mesh-build-go's single manifest ask writes and clears its first-view-unanswered
finding through the writer with the contract's Finding (A-28). mesh-config and mesh-vault's own followers (A-28) are not
yet moved onto a views.Reader: they do not compile at mesh-sdk 24f876a.
Builds: design 47 G §687, §689, A-17, A-22, A-26, A-28
Claude-Session: https://claude.ai/code/session_01JG73KEq4zoDfM9WtYpcQgZ
artifact-store-tools (image-missing, sweep-refused), distribution (program and tools bundle, one writer each over its own
kinds), gitea's package-registry (and the token-refused key the TypeScript bundle watches) and mesh-config keep their
findings open and write and clear them through the generic writer; mesh-vault states its assignments view's. A part of an
id told apart by a ref is in the id (<id>.<ref>), each escaped; the writer starts once Serve has the runtime's channel; the
removed findings helpers and resolved words go.
Builds: design 47 G §686, §687, §689, A-3, A-14, A-16, A-22
Claude-Session: https://claude.ai/code/session_01JG73KEq4zoDfM9WtYpcQgZ
merges and source stay in the old state section; pulls and asks wait for the removal of the merge states.
Builds: design 47 G §613, A-8, A-20 (G-17, G-20)
Claude-Session: https://claude.ai/code/session_01JG73KEq4zoDfM9WtYpcQgZ
nfs-server, minio object-store, claude-code, mesh-watcher, mesh-provisioner and mesh-build-go hand their views reader
views.WriterFindings over views.NewStdioWriter and start it; the removed meshconditionsclient.FindingWriter goes.
Builds: design 47 G §686, §687, §689, A-3, A-16, A-17
Claude-Session: https://claude.ai/code/session_01JG73KEq4zoDfM9WtYpcQgZ
GATE17: record.go Remove asks buscalls.LedgerRemove, recordfill.go reads buscalls.Fill, the manifest names ledger-upkeep-calls.
GATE11: the verb settle, its ask, EvSettled and the Settle field leave with the merge states.
Builds: design 47 G §587, §431
Claude-Session: https://claude.ai/code/session_01JG73KEq4zoDfM9WtYpcQgZ
The gate's source value is the contract's SourceByBranchValue and its key SourceKey(branch); the verb takes an optional
branch and answers branches alone when given none. The whole-module package does not compile on the join for GATE11
(settle) and GATE17 (ledger fill and remove), so the unit suite was run with those five sites stubbed in a scratch edit,
never committed: green apart from the stubbed removal case.
Builds: design 47 G §690, §614, §15
Claude-Session: https://claude.ai/code/session_01JG73KEq4zoDfM9WtYpcQgZ
Reverts bbd9f638, c2e017a4 and b6884796; cmd/mesh-nats-snapshot/restore.go is as before them. The manifest test
asserts the claim serves only the reading verbs and the keeper's state is asked, requests, spent.
Builds: design 47 G §554, §555
Claude-Session: https://claude.ai/code/session_01JG73KEq4zoDfM9WtYpcQgZ
The gate reads its records' ledger fill before it opens a record; its manifest asked no
capability, so it could run beside a runner answering neither. It now asks ledger-calls and
ledger-fill, the fill a capability of its own.
Claude-Session: https://claude.ai/code/session_01JG73KEq4zoDfM9WtYpcQgZ
(Joined from c20044a2.)
Builds: design 47 G §181, §462, §488 (GATE12)
Claude-Session: https://claude.ai/code/session_01JG73KEq4zoDfM9WtYpcQgZ
ADR 0659 §4 kept three description reads that shape words only until they move to the bus: the
issue numbers the ask names, the author a stalled line names and the notice that after: lines are
not read. The ask now names the issues linked to each member's pull request, read once per head
through the seat issue-tracker's list (a tracker that does not answer is said, never waited on);
the author is the forge's pull request record's, from the seat git's events; the after: notice is
gone. The gate's forms of a pull request and a delivery hold no description field.
Claude-Session: https://claude.ai/code/session_01JG73KEq4zoDfM9WtYpcQgZ
(Joined from ddd93a78.)
Builds: design 47 G §50, §143 (GATE15)
Claude-Session: https://claude.ai/code/session_01JG73KEq4zoDfM9WtYpcQgZ
Design 47: a store filling is said before it is full. The gate reads its ledger's fill through the
tool runner's mesh/ledger.fill, says filling of the seat mesh-delivery when a record takes the ledger
past 80 % of its cap, refuses the next record when none fits and says full naming it, and keeps the
merge waiting for room on its delivery (held_full), so a restarted gate judges alike. Once the
ledger is below 80 % again, after the retention's removals, it says room and hands the merge.
gate_records says the fill against its cap, or why it is unread.
Claude-Session: https://claude.ai/code/session_01JG73KEq4zoDfM9WtYpcQgZ
(Joined from 061be00d.)
Builds: design 47 G §181 (GATE12)
Claude-Session: https://claude.ai/code/session_01JG73KEq4zoDfM9WtYpcQgZ
ADR 0661: the check gates the ask, in one place (askWaits); once the Approve covers
the heads (approvedWaits) no verdict, carry, composed check or required status is
judged, an Approve given before a trunk move merges onto the new trunk, and a
failure after the Approve is said once on the pull request and as a stalled line
naming the merge it follows. A check failing before the answer withdraws the ask.
Claude-Session: https://claude.ai/code/session_01JG73KEq4zoDfM9WtYpcQgZ
(Joined from 43696644.)
Builds: design 47 G §278, test 52 (GATE10)
Claude-Session: https://claude.ai/code/session_01JG73KEq4zoDfM9WtYpcQgZ
The module's own manifest is embedded (package at the module's root); an event the gate hears from another module is
the consumes entry with that event's name after its owner. No new seat, no second publisher on the seat git (the
lead's ruling). Rows in the manifest test, which also reads the gate's sources for those names.
Claude-Session: https://claude.ai/code/session_01JG73KEq4zoDfM9WtYpcQgZ
(Joined from 2de87fd8.)
Builds: design 47 G §7, §146 (GATE16); the bridge's name stays in code for the old merge follower (eb131beb left out, G §167, §542)
Claude-Session: https://claude.ai/code/session_01JG73KEq4zoDfM9WtYpcQgZ
Every key of owed, and the member merging in every key of merges, is said as owed under
meshdeliveryclient.SwitchedActID(state, key, revision), at every switch, the same act each time. The switch reads
the old states through one reader (OldState). Rows in the switch's test.
Claude-Session: https://claude.ai/code/session_01JG73KEq4zoDfM9WtYpcQgZ
(Joined from dd0c4294.)
Builds: design 47 G §169 (GATE5, GATE6); only the owed carry: the member merging in the old state merges is not carried as a merge act (G §167, §542)
Claude-Session: https://claude.ai/code/session_01JG73KEq4zoDfM9WtYpcQgZ
At the gate's start every unanswered act is said again under the same id; an outcome is written into the record that
holds the pull request, then answered; an outcome other than performed is said; the checker's are not the gate's.
Needs the forge bridge's forgemsg owed.ended (feat/47-G-forge).
Claude-Session: https://claude.ai/code/session_01JG73KEq4zoDfM9WtYpcQgZ
(Joined from cf3715ea.)
Builds: design 47 G §162, §163, §164, §165, §166 (GATE5)
Claude-Session: https://claude.ai/code/session_01JG73KEq4zoDfM9WtYpcQgZ
The base's protection and required contexts, the forge's conflict word and the merging
account's write access come from the pull answer's merging (issues 1275, 1945; ADR 0441 §1);
a required context not a success at the commit blocks.
Claude-Session: https://claude.ai/code/session_01JG73KEq4zoDfM9WtYpcQgZ
(Joined from 13e12d2c.)
Builds: design 47 G §252, §310, §312 (GATE14)
Claude-Session: https://claude.ai/code/session_01JG73KEq4zoDfM9WtYpcQgZ
The gate named the forge module's tools and events and guessed their answers' shape (operator (d), F14).
Its reads now are the seat git's verbs pull, statuses, open-pulls and contains, typed in the seat's
contract, a failed read an error and never none; its forge events are the seat's pull.updated,
pull.merged, pull.closed, forge.pull, forge.status and repo.writable, read into the gate's own forms
in one place (gitseat.go). What the forge would merge now (the base's protection, its conflict word,
whether the merging account writes) has no verb of the seat yet: it stays the one read of the
forge's statuses tool, named as such (mergeStateTool), until the seat carries it.
The bus tests' runtime now serves the gate's ledger, so the two whole-bundle tests that failed
since W11.1 pass again.
Claude-Session: https://claude.ai/code/session_01JG73KEq4zoDfM9WtYpcQgZ
(Joined from 14338514.)
Builds: design 47 G §252, §310, §312, test 51 (GATE14)
Claude-Session: https://claude.ai/code/session_01JG73KEq4zoDfM9WtYpcQgZ
Design 47 (The switch carries today's state in, once): at the gate's start, before any event is
taken, each group in flight becomes a record keyed by its milestone and its heads (the heads its
merge was approved at when merges holds one), with that merge as its group state; a second switch
finds each record there and opens none, so a restart or a delivery of milestone G again after a
revert asks nothing again. From then on every merge the gate hands is written into its record too,
so the record's outcomes are found by its members' merge commits.
Left for W11.17 and W11.14: the merge handed as an owed merge-pull instead of the merges key, the
owed keys carried, and merges and turns no longer declared.
Claude-Session: https://claude.ai/code/session_01JG73KEq4zoDfM9WtYpcQgZ
(Joined from e9415814.)
Builds: design 47 G §156, §220 (GATE4)
Claude-Session: https://claude.ai/code/session_01JG73KEq4zoDfM9WtYpcQgZ
Design 47 (The delivery record): the outcomes arrive as events through the module's durable
consumer, each applied once and only from its owner's account. A family per step event names the
pattern the manifest consumes, its owner and whether its subject names a node; an event is applied
only when its key names its owner and, for a node's event, its x-node is the node its subject
names. It is taken once its outcome is in every open record holding it (by a member's merge
commit, or by the entry of the record's graph difference), so one the ledger did not take is
offered again; an unreadable one is said and taken. The indexer's take-in.taken is the first
family; each other step's event joins the table with its owner's contract.
Claude-Session: https://claude.ai/code/session_01JG73KEq4zoDfM9WtYpcQgZ
(Joined from 4c4d1239.)
Builds: design 47 G §34, §141 to §146 (GATE2)
Claude-Session: https://claude.ai/code/session_01JG73KEq4zoDfM9WtYpcQgZ
ADR 0656 §7 and design 47 (The delivery record): the gate asks the bus module for a ledger
`records` (template ledger, capped at 768 MiB) with two types: a record, keyed by its milestone and
its heads, whose operations open it, write its group state, trace, outcomes, nodes away, pinned
revisions and end, and remove it; and a month's list of the records opened, since a bundle cannot
list a ledger's instances. An open of heads already recorded leaves the record as it is, an outcome
applied twice writes nothing, and each open removes the records past the setting record-retention
(90 days by default), naming each revision they pinned. The layout is the gate's own (design 20).
Claude-Session: https://claude.ai/code/session_01JG73KEq4zoDfM9WtYpcQgZ
(Joined from 84eda463.)
Builds: design 47 G §141, §142, §143, §144, §145, §146, §155, §34 (GATE1)
Claude-Session: https://claude.ai/code/session_01JG73KEq4zoDfM9WtYpcQgZ
Design 47 lines 1888-1906 at b229f5c4 (test 60): the verb sits in every build seat's definition.
Its judgement (what may be kept, the kept builds a node holds, the three refusals) is shared code of
one domain's modules, so it is the domain SDK's (sdk/buildstep KeptArtifact, ADR 0634 §2):
mesh-build-go keeps its store, sweep and tools and drops its own copy of the judgement and its table;
mesh-build-node and mesh-build-image keep no build and answer over none, so a module claiming no
oci-artifact-registry is refused and the holder is answered not kept. Their manifests serve the verb
and invoke seat:mesh-catalog.module, which the judgement asks.
Claude-Session: https://claude.ai/code/session_01JG73KEq4zoDfM9WtYpcQgZ
(Joined from 42d364fe.)
Builds: design 47 G §271, §464 (BLD9)
Claude-Session: https://claude.ai/code/session_01JG73KEq4zoDfM9WtYpcQgZ
Builds design 47 *The artifact store's holder takes its own new build even
when its seat is the one broken* (lines ~1811-1830, test 60): a trunk build of
a module claiming oci-artifact-registry that the seat refuses or does not
answer is kept as an object by digest and answered kept (BuildAnswer.Kept);
kept-artifact answers its store and object, refusing a module that claims no
oci-artifact-registry, one not kept here and one under another digest. The
store is the template objects (ADR 0666 §4: a part outside the foundation keeps
its data in ADR 0595's templates; design 47's "old objects form" contradicts
it for a non-foundation part, reported). Whether a module claims the seat is
asked of the seat mesh-catalog's verb module (ADR 0596). Callers are left to
the seat definition and the grants part (ADR 0666 §20). build_go_kept reads the
kept builds, build_go_kept_drop drops one (ADR 0666 §25).
Claude-Session: https://claude.ai/code/session_01JG73KEq4zoDfM9WtYpcQgZ
(Joined from 5e1cdade.)
Builds: design 47 G §271, test 60 (BLD9)
Claude-Session: https://claude.ai/code/session_01JG73KEq4zoDfM9WtYpcQgZ
A test step runs only the packages with no kept result for their digest, and the runner makes
that carry in the checker's place (ADR 0666 section 8; design 47 *Carried, not run* line 593 at
hq b229f5c4): the kept results reach the step in its ask (StepRun.kept, mesh-delivery
dc0df1b5), so the module's reads lose mesh-delivery-checker.kept and the lookup of the
checker's kept bucket goes. The carried and kept packages are in the step's end, on its step
key.
Milestone 47/G, research 086 area C (K1).
Claude-Session: https://claude.ai/code/session_01JG73KEq4zoDfM9WtYpcQgZ
(Joined from 1c991294.)
Builds: design 47 G §25, §27, §66, §416 (BLD5)
Claude-Session: https://claude.ai/code/session_01JG73KEq4zoDfM9WtYpcQgZ
Lint is a pipeline step of its own, served by the build seat (ADR 0666 section 7; design 47
*The steps and who serves them* line 571 at hq b229f5c4), so the unit level runs tests alone
and the lint step runs gofmt and go vet, with every tag the files name, over each go.mod
directory of the artifact's module directory; a module directory holding no go.mod has
nothing to lint (prepare finding F12).
Milestone 47/G, research 086 area A.
Claude-Session: https://claude.ai/code/session_01JG73KEq4zoDfM9WtYpcQgZ
(Joined from a46e01a2.)
Builds: design 47 G §328, §417 (BLD6)
Claude-Session: https://claude.ai/code/session_01JG73KEq4zoDfM9WtYpcQgZ
The runner asks a lint or test step of an artifact as the build seat's accepted kind step, and
hears its end as the seat's step-ended on the node that ran it (ADR 0666 sections 7, 8, 21;
design 47 *The steps and who serves them* line 571 and *Asked on a work queue, ended by an event*
lines 583-590 at hq b229f5c4). The build modules' shared package runs a test step as a check of
its one level, carries the packages whose digest the ask names kept, keeps what it passed and
says the counts; a level with no tests ends nothing to run, a source fault could not run, and a
transport fault or an end that could not be said is handed back unsaid, so the work queue gives
the step again and it ends once under its ask's id. The neutral forms are the domain SDK's
(sdk/buildstep pipelinestep.go, mesh-delivery dc0df1b5); the Go module says the end through the
runtime's seat publish.
Milestone 47/G, research 086 area A.
Claude-Session: https://claude.ai/code/session_01JG73KEq4zoDfM9WtYpcQgZ
(Joined from 72fee3c6.)
Builds: design 47 G §22, §24, §86, §116, §140 (BLD3, BLD4)
Claude-Session: https://claude.ai/code/session_01JG73KEq4zoDfM9WtYpcQgZ
The container image and OCI artifact seats' holder asks the seat mesh-builds' verb reserved in place of builder_versions; the manifest
invokes it and takes no builder package.
Builds: design 47 G §381, §478, §504
Claude-Session: https://claude.ai/code/session_01JG73KEq4zoDfM9WtYpcQgZ
The holder asks the seat mesh-builds' verb reserved with the publish's name and version in place of builder_versions, and stores
only a version reserved for an artifact of the publish's digest; the manifest invokes the verb and takes no builder package.
Builds: design 47 G §381, §478, §504
Claude-Session: https://claude.ai/code/session_01JG73KEq4zoDfM9WtYpcQgZ
merge-pull refuses a head at whose fingerprint a run failed a required step (policy-failed, naming run, module, step), from the
module's own state failed-steps kept from step-ended, with the seen keys of every step end and the deletion once no watched
branch's head nor open pull request's head names a commit. The two tools read it with its fill and drop a key after the hand act.
Builds: design 47 G §39, §83, §84, §85, §135, §475, §487, §548
Claude-Session: https://claude.ai/code/session_01JG73KEq4zoDfM9WtYpcQgZ
The fixed Observers list and the engineModule constant are removed; sources are the modules whose catalog entry declares the
finding events, a module in no branch's revision is cleared (module left the catalog), a stopped module no longer placed is
cleared (module left the node), Keeper.Holder is derived from the entry's claims. The per-branch catalog followers in main.go
wait on a design question (how a part adopts one catalog view per registered branch); the whole integration test is red until then.
Builds: design 47 G §365, §196, §200, §364, §384
Claude-Session: https://claude.ai/code/session_01JG73KEq4zoDfM9WtYpcQgZ
Sends reads a claim of a seat whose definition has a verb or an event, or an event of its own; AsksOf reads
an entry's bus-state asks and claims. The follower's branch and the wiring wait on the design's answer.
Builds: design 47 G §344, §433
Claude-Session: https://claude.ai/code/session_01JG73KEq4zoDfM9WtYpcQgZ
Each of the six declares the stream <module>-findings (finding-written.*, finding-cleared.*,
findings-stated.*, 7 days, 16 MiB) and consumes mesh-conditions.findings-asked; the manifest
tests that hold consumes say it. The other observers' manifests are written by their own families.
Builds: design 47 G §387, §383
Claude-Session: https://claude.ai/code/session_01JG73KEq4zoDfM9WtYpcQgZ
An observer on a machine away is not named in findings-unstated; a machine back has its
observers asked again (mesh-watchdogs machine-away and machine-back in consumes).
Builds: design 47 G §389
Claude-Session: https://claude.ai/code/session_01JG73KEq4zoDfM9WtYpcQgZ
resync writes the hand-act log, says findings-asked for every observer or the one named and
settles what it holds; an observer silent five minutes after an ask (bounds state, key time
to live), its machine not away, is the keeper's finding findings-unstated of scope module,
cleared by its next findings-stated; conditions_findings_views answers each observer and node.
The nodes asked are those heard (the kept list of COND2 is stopped on a question).
Builds: design 47 G §390
Claude-Session: https://claude.ai/code/session_01JG73KEq4zoDfM9WtYpcQgZ
Verbs snapshot-kept and restore-kept (with aside-only and discard-aside) on the seat mesh-broker, handed to the
keeper; taken over from the join's 0480e16f and c4e88eaa, without the verbs aside and discard-aside and the tool
nats_kept, which no design line gives.
Builds: design 47 G §296, §297, §302, §337, §405
Claude-Session: https://claude.ai/code/session_01JG73KEq4zoDfM9WtYpcQgZ
findings-stated raises what is new, keeps each pending key unchanged, clears every other
condition of the observer on that node with its reason, clears a pending key past its bound
(bounds state, key time to live), and for an observer held once takes the holder's set and
later events only. The holder lookup is a hook (Keeper.Holder) that COND6 wires from the
catalog; the view is kept in memory until COND2 keeps findings-views.
Builds: design 47 G §384
Claude-Session: https://claude.ai/code/session_01JG73KEq4zoDfM9WtYpcQgZ
The store is made by Keeper.Sends, kept when a verdict asks nothing of it, and a full one says `full`.
Which module sends is NATS3's input (the catalog half); this makes the store.
Builds: design 47 G §59, §137, §344, §477
Claude-Session: https://claude.ai/code/session_01JG73KEq4zoDfM9WtYpcQgZ
The keeper takes finding-written and finding-cleared from each observer's stream
(`*.finding-written`, `*.finding-cleared`, `*.findings-stated` in consumes); its reads of
every `findings`, the views.findings consumes and the read_findings invokes go.
node-engine's name is a local constant since mesh-sdk's buscompose lost it.
Builds: design 47 G §383, §386
Claude-Session: https://claude.ai/code/session_01JG73KEq4zoDfM9WtYpcQgZ
The base "main" fallbacks of the open-pulls read and the announcer are gone (the indexer's module name and tool went with GIT10).
The manifest's version is MAN1's, written with the fill's source it needs.
Builds: design 47 G §7, §428
Claude-Session: https://claude.ai/code/session_01JG73KEq4zoDfM9WtYpcQgZ
The config value callers, the package seats' list of callers with its watch, and the seat git's caller judgement are gone:
the grants part grants by each definition's callers and the bus refuses the rest.
Builds: design 47 G §68, §127, §313, §321
Claude-Session: https://claude.ai/code/session_01JG73KEq4zoDfM9WtYpcQgZ
A tag at that commit is answered performed; at another commit, refused (a tag is never moved). The builder alone calls it:
the bus refuses any other account from the definition's callers.
Builds: design 47 G §370
Claude-Session: https://claude.ai/code/session_01JG73KEq4zoDfM9WtYpcQgZ
merge-pull answers its refused-kind (head-moved, unprotected, git-refused, no-registrations); an already merged pull request at
that head is answered performed. The failed-steps refusal (policy-failed) is GIT7, stopped on a design question.
GIT9: registrations.ts learns the first registrations from the seat mesh-catalog's verb registrations at the start and its event
registration-changed; the code artifact needs mesh-delivery-assigner:program.
Builds: design 47 G §38, §129, §314, §95, §96, §130, §353, §432
Claude-Session: https://claude.ai/code/session_01JG73KEq4zoDfM9WtYpcQgZ
The four share files (index.ts, module.json, guard.test.ts), so they land as one commit.
GIT1: seat.ts serves pull, open-pulls, tags, trees, file, contains, statuses, status, comment, view, note, merge-pull
as tools on the seat, each answering the contract's shape and an error of kind refused, not-found or failed; a comment,
view or note whose act id is held is answered held (note matched by id). merge-pull merges the head the caller names
(its from-G judgements are GIT6, GIT7, GIT9).
GIT2: seatevents.ts says pull-updated, pull-merged, pull-closed, forge-pull, forge-status, forge-delete and repo-writable on
mesh.seat.git.event.<event>; the module's emits drop them.
GIT3: merge.ts, check.ts, the acting tools, the owed watch, the gate's source read and check-gates-merge are gone.
GIT10: catalog.ts follows the seat mesh-catalog (revision, catalog-revision); no write is granted before the first view.
Builds: design 47 G §310, §311, §312, §313, §314, §163, §317, §278, §321, §196
Claude-Session: https://claude.ai/code/session_01JG73KEq4zoDfM9WtYpcQgZ
Node answers the type-check as lint always, unit where a package has a test
script and integration where it has test:integration, kept in detected-steps by
fingerprint as the Go module keeps it; the image answers none. The manifests
serve it. The step ask's end (step-ended) of BLD7 and BLD8 waits for BLD3 and
the build-step package of mesh-delivery-sdk.
Builds: design 47 G §22, §307, §326, §328
Claude-Session: https://claude.ai/code/session_01JG73KEq4zoDfM9WtYpcQgZ
Part of BLD15: the manifest's stream mesh-build-go-findings (the three
finding events, 7 days, 16 MiB) and the consume of mesh-conditions.findings-asked.
The module's writing of findings through mesh-sdk's finding writer waits for
the writer (the writer's API is an open design question, D-16).
Builds: design 47 G §383, §384, §387
Claude-Session: https://claude.ai/code/session_01JG73KEq4zoDfM9WtYpcQgZ
The Go build module asks the input verb in the ask (the checker's, the
indexer's mesh-catalog input or the assigner's mesh-wants input) and its
manifest invokes the two seats' verbs. The bus refuses the other accounts
(grants of the contract's callers). The step key keeps the contract's form
<node>.<id>: no asker part is named there (question to the lead).
Builds: design 47 G §125, §206, §338, §354
Claude-Session: https://claude.ai/code/session_01JG73KEq4zoDfM9WtYpcQgZ
The Go compile, the Node tsc/esbuild/npm and the image's buildah bud get the
build argument MESH_VERSION from the version the build ask carries. The late
publish on version.numbered goes (code, tests, the consumes of the Go and
image manifests). The image's label and tag and a program's stamp are open
(question to the lead): the design names no key or symbol.
Builds: design 47 G §349, §370, §413
Claude-Session: https://claude.ai/code/session_01JG73KEq4zoDfM9WtYpcQgZ
Part of BLD2 (Go module): the state detected-steps asked by the template
state, one key per artifact, answered for the same fingerprint, replaced by a
newer one and by again. The drop of a key and the Node and image modules wait
(drop: question to the lead; the others follow in BLD7 and BLD8).
Builds: design 47 G §326, §414
Claude-Session: https://claude.ai/code/session_01JG73KEq4zoDfM9WtYpcQgZ
The verb fetches the artifact's tree and answers lint always, unit and
integration steps by the packages holding test files (integration by the tag),
a level with no test left out; the manifest serves it.
Builds: design 47 G §22, §307, §28
Claude-Session: https://claude.ai/code/session_01JG73KEq4zoDfM9WtYpcQgZ
An event whose subject carries parts is published by a function taking each part and heard by a filter with a * per
part; an event naming keep gets the stream SEAT_<seat>_<event>, in Go and in the TypeScript contract. The reader of the
sdk's reader release reads acts (every, a language, kind:image), keep as an object and callers as a list.
Builds: design 47 G §89, §203, §238, §260, §324, §331, §396, §397
Claude-Session: https://claude.ai/code/session_01JG73KEq4zoDfM9WtYpcQgZ
A seat's npm package holds index.js, index.d.ts and parts/<part>.js with .d.ts, with no TypeScript source and no
tsconfig.json, so a TypeScript module imports it with no build step and no tsc of the contract.
Builds: design 47 G §341
Claude-Session: https://claude.ai/code/session_01JG73KEq4zoDfM9WtYpcQgZ
The generator checks each verb's and accepted kind's callers in a definition and a part before reading it, naming the
file, the verb and the entry; text is refused as one entry; the contracts it writes carry no caller check.
Builds: design 47 G §336, §339, §392
Claude-Session: https://claude.ai/code/session_01JG73KEq4zoDfM9WtYpcQgZ
mesh/check checking ae056a44dd72 on 9dfb2eae0eb3 | novox/mesh-catalog#656 in group fix/1966-bindings-first-view
Held.binding is the entry's plan binding, the value its created and recreated
carry, present when the bus holds the entry's objects and absent on one it does
not (waiting, refused). Table rows in the reading verbs' test: the binding of a
held entry equals the one its created carries; a waiting entry has none.
Builds: design 25 bindings-first-view §2, §7 (hq d0760629, PR 543)
Claude-Session: https://claude.ai/code/session_01JG73KEq4zoDfM9WtYpcQgZ
go mod download loads only the module graph a go.mod names; a vet, a test or a build then loads
packages shut off from the network and failed far from the fetch, in fewer words (the builder's
"pull_drop.go:11:2: module lookup disabled by GOPROXY=off"). After each directory's download the
fetch loads its packages with their tests shut off and fails naming each package whose load
would reach the network. The graph of a sibling laid beside is loaded once with the network
first, so a go.mod the composer's build reads and the graph's download skipped is fetched, then
proven the same way.
1966/AC178-2: a row of TestAVerifyThatFailsEndsTheStepNamingTheDirectoryAndTheLine; the beside
case in TestTheFetchDownloadsTheGraphOfASiblingModuleLaidWithTheTreesItRequires.
Claude-Session: https://claude.ai/code/session_01JG73KEq4zoDfM9WtYpcQgZ
ADR 0573 §2 builds every module against the latest published version of each mesh package it
uses, whatever its go.mod names. The fetch took versions only for packages the directory's own
go.mod names, so a package only a module it replaces by a directory requires (mesh-build-image
replacing buildstep, which imports the node-build-go seat package) was never taken: go mod
download passed and every shut-off step looked it up on the network. Now the requirements of
each module the directory replaces by a directory count too, taken the same way (the ask, its
repository's asked head, else the latest) and said as named by that directory.
1966/AC178-1: rows of TestARequirementAtTheZeroVersionIsTakenFromTheAskOrTheTrunk.
Claude-Session: https://claude.ai/code/session_01JG73KEq4zoDfM9WtYpcQgZ
Since buildstep's drop verb imports the node-build-go seat package, both modules' shut-off vet,
test and build looked it up on the network: their go.mod did not name it, so the fetch took no
version for it. go mod tidy adds it as an indirect requirement.
Claude-Session: https://claude.ai/code/session_01JG73KEq4zoDfM9WtYpcQgZ
mesh/check checking c39503e3227c on c4463ba47fcf | novox/mesh-catalog#650 in group fix/1966-seat-list-unread
claude_code_seat_list_again answered person.<name> and the healers, but every
call through the mesh MCP server is proved as its machine's tool runner
(issue 627); node-tools now stands for that session (issue 1966).
Claude-Session: https://claude.ai/code/session_01JG73KEq4zoDfM9WtYpcQgZ
ace's node-engine wrote seat-list-unread in its state findings, and no
condition rose: the keeper's observers did not name the node-engine. It now
reads mesh-host.findings, so ADR 0636 §1's condition is raised and clears at
the first full read (issue 1966).
Claude-Session: https://claude.ai/code/session_01JG73KEq4zoDfM9WtYpcQgZ
GOPROXY joined the Go seat, the public proxy and direct with commas, so go went on only after a 404 or 410: a contracts
package the seat does not hold, and a public proxy a node cannot reach, ended the fetch before direct. It joins them
with | now, so any failure goes on to the next. go words a proxy it cannot reach as "invalid version", which the pin
path read as an unknown revision and labelled "on no pull request head"; a network failure is never that now
(network is unreachable and no route to host are network failures).
Claude-Session: https://claude.ai/code/session_01JG73KEq4zoDfM9WtYpcQgZ
A lost event between a shared stream and a module's consumer could not be located: no tool read a consumer's delivered and acknowledged sequence or a stream's messages for one subject. nats_consumer reads from the monitoring document; nats_stream_messages asks STREAM.MSG.GET on the module's own credential, newest first and bounded.
Claude-Session: https://claude.ai/code/session_01JG73KEq4zoDfM9WtYpcQgZ
hand-acts read the whole log (84k entries, 47 MB) on every call and could never answer inside the call bound or
the bus payload. It now reads from the newest entry backwards, page by page, and stops at the window's start, at
900 KiB of acts or at 20 s, saying complete: false, cut and why. warrants is bounded the same way. mesh-healers'
start-up seed takes a partial answer and logs it, instead of retrying forever.
Quick fix: this (mesh-catalog only, about 1.5 h). Thorough fix: also the tool runner never freezing on an oversized
bundle line and logging unanswered calls (mesh-tools, a separate fix by another developer, lead's ruling Q3), and an
SDK-side bound. Chosen here: quick for this module, the thorough part running beside it; the byte cap is the
module's own (lead's ruling Q1, a `limit` argument is a later note).
Claude-Session: https://claude.ai/code/session_01NcbgfFfQ4Z9UE5jK32PmE5
mesh/check checking 4e2a1a5a1f58 on 503411e6289b | novox/mesh-catalog#640 in group fix/1966-checker-compositions-objects
The checker sends a large node's request packed, as it passes the bus's message limit; compose reads either form through
ComposeJob.RequestBytes. Only mesh-build-go serves compose. Delivered as a group with the mesh-delivery change of the same
branch name: an older mesh-build-go refuses a packed ask naming that it holds no request object.
Claude-Session: https://claude.ai/code/session_01NcbgfFfQ4Z9UE5jK32PmE5
mesh/check checking 80186d079808 on 6a24e4d0c77e | novox/mesh-catalog#639 in group fix/1966-forge-made-merge
Merges made on the forge directly since 2026-11-09 07:04 UTC have no gate merge record,
and the indexer holds them; an empty merge through the gate covers this repository's head.
Second of its kind: the group's mesh-delivery member removes the need (the forge bridge
reads a forge-made merge's pull request from the forge).
Claude-Session: https://claude.ai/code/session_01NcbgfFSnn24M3GcqttoU7i
Not read is not unneeded on the newest scope too (issue 940's rule): a module whose newest composition is
unpublished or refused is held at its newest build, said in provisioner_plans newest-held and bounded on the health.
A retiring key needed again cancels its ask on operator-channel (accept.cancel), retried until the router takes it.
Claude-Session: https://claude.ai/code/session_01NcbgfFSnn24M3GcqttoU7i
The composer finds a listen's machine port in a container mapping that names
it on either side (MachineSide). The program's listen 5000 matched the
registry container's inner 5000 in 127.0.0.1:5001:5000, so the store's address
became the container's loopback side and the program's own port was never
opened: the move to 83cd1a3d on 2026-11-08 23:25 UTC cut the artifact store off
(issue 1966 note 109). The program now binds 5100, where every machine fetches
(the artifact-store alias and the seats' reach agree), and the container serves
5001 inside and out. A row holds that no container names 5100.
Undo: revert this commit.
Claude-Session: https://claude.ai/code/session_01NcbgfFSnn24M3GcqttoU7i
nats_keeper_health found the keeper's consumer novox_nats_keeper in /jsz, but the read it shares with the
broker verbs asked /jsz?accounts=1&streams=1&config=1, without consumers=1, so keeperPending always read -1
and the keeper was "unhealthy: what waits on its consumer cannot be read from the bus's monitoring" for
ever (2026-11-08, issue 1966).
Claude-Session: https://claude.ai/code/session_01NcbgfFfQ4Z9UE5jK32PmE5
mesh-knowledge's stream reviews says max-age-days "none", which the indexer's judge accepts
(judge/busstate.go noAge), but the keeper refused it ("not a whole number"), so the stream was
never made and the keeper stayed unhealthy (2026-11-08, issue 1966). Quick fix; the thorough one,
one shared reader of the template options in mesh-sdk catalogue for the judge and the keeper, is
noted on issue 1966.
Claude-Session: https://claude.ai/code/session_01NcbgfFfQ4Z9UE5jK32PmE5
The two server lines were literals in the template, so changing an upstream needed a build. They are now the config keys upstream-resolver-first (1.1.1.1) and upstream-resolver-second (8.8.8.8), placed with ${config:...}; the service already restarts on its config file. no-resolv and every other line stay.
Claude-Session: https://claude.ai/code/session_01NcbgfFfQ4Z9UE5jK32PmE5
The caps were literals in server.env and the container named no restart-on, so a changed env-file never reached the running forge. They are now three gitea config values placed with ${config:...}, and the server container restarts on server-env.
Claude-Session: https://claude.ai/code/session_01NcbgfFfQ4Z9UE5jK32PmE5
mesh/check failed (3): ace: the check of novox/mesh-catalog#632 failed: downloading the asked version 68581…; and 2 more | b4192c525e7e on 61737962fcb6
Kind: short package names (builds, sshkeys, forgemsg, backupgood,
diskreadings, grantsstate, busgrants) in `packages`, which the indexer's
judge refuses by design 32's rule (a Go package is named by its go.mod's
module path). This held mesh-delivery-gate, the build modules, mounts and
restic. Each key is now the import path the module's code uses, with its
series kept; the manifest tests read the new keys, and the host/machine-name
word tests in mounts, nfs-server, mesh-build-node and mesh-delivery-gate
look past the path's `git.novox.be/novox/` prefix.
Quick fix: only the gate's manifest. Thorough fix (chosen): all 15, since
the same refusal holds each. Follow-up for the checker: the candidate
verdict lets such a name through (J7(3)), so the merge check did not stop it.
Claude-Session: https://claude.ai/code/session_01NcbgfFfQ4Z9UE5jK32PmE5
The operator, 2026-11-07: remove any gate that reads a description. An unreviewed pull request whose
other waits are met is asked, and the ask names no review line.
Claude-Session: https://claude.ai/code/session_01NcbgfFfQ4Z9UE5jK32PmE5
With check-gates-merge false (the configuration seat's value) the gate still held the Approve until the check had
a verdict (verdictWaits), so a one-line recovery waited half an hour for a whole-suite check it does not gate. The
operator, 2026-11-07: 'we aren't running any checks anymore, we're delivering'; 'disable the check'. The Approve is
now asked once the delivery is reviewed and ready; a later verdict is still shown and asks the pending Approve again
in its words. Tests follow: asked at once with no verdict or a running check, a group too, a re-pushed member too.
Merged by the session as a recovery fix (instruction recovery-fixes-merge-and-tell), the operator told on Telegram.
Claude-Session: https://claude.ai/code/session_01NcbgfFSnn24M3GcqttoU7i
mesh/check checking 83732dcd53d8 on 9a3356bd38b1 | novox/mesh-catalog#627 in group fix/1966-bootstrap-no-packages
The running builder predates Go package artifacts ("refused kind: package of language go ... no build module
builds it"), so it can build neither the new builder, the grants module, the Go build module nor the checker:
each new part is needed before the others can be built or delivered. For the bootstrap of the registry
delivery, at the operator's word (19:4x, "do the temporary change"), the package artifacts leave the manifests of
the parts that run; the contracts and SDK modules, which run on no node, keep theirs. They come back in one change
once the new builder runs. Quick vs thorough: quick; the thorough fix is 47/G's tier order (issue 1921 note 80).
Claude-Session: https://claude.ai/code/session_01NcbgfFSnn24M3GcqttoU7i
mesh-catalog #625 offered artifact-store again with serves port 5000, copied from the manifest before the
registry milestone. The running node-engines turn that port through the ports the node assigned the module; with
the milestone's manifest the assignment no longer turns 5000 into 5100, where the store's container has published
since 2026-10-04 (docker_ports on novox: 0.0.0.0:5100->5000). Since #625 was taken in, novox fetches from :5000
and is refused (node-engine journal 18:54: connection refused), while the store's log shows fetches on :5100
until 18:47:52. The alias now names 5100. Removed with the alias once every node-engine runs the milestone's build.
Claude-Session: https://claude.ai/code/session_01NcbgfFSnn24M3GcqttoU7i
The registry milestone renamed what distribution offers (container-image-registry, oci-artifact-registry,
mesh-artifact-store), as ADR 0632 section 3 decides. The node-engines running on every node are the ones
from before it: they find the store through the provision artifact-store, so they fetch no new build,
including their own successor, and keep running what they ran. ADR 0632 section 2's guarantee (no node loses
the store while its engine reads it the old way) lives in the new indexer's hold, which the running indexer
predates. So this goes against ADR 0632 section 3 for a limited time, at the operator's word (bootstrap of
2026-11-07, no new record): offered again beside the new roles until every node-engine runs the registry
milestone's build, then removed in the next change to distribution.
Claude-Session: https://claude.ai/code/session_01NcbgfFSnn24M3GcqttoU7i
mesh/check failed: shanks: the check of novox/mesh-catalog#624 failed: downloading the asked version 54aff0a3d9201e3df2… | 77b30bd2966c on f2b332de73d6
The group check of 8a80e8e86858 laid every Go module beside the check, and one of them (mesh-delivery's
mesh-grants, which now requires disk-load's diskreadings) required a catalogue module the catalogue under check
never lays beside itself: Go looked its zero version up and the composer did not build.
Claude-Session: https://claude.ai/code/session_01NcbgfFSnn24M3GcqttoU7i
mesh/delivery-group group feat/registry-milestone rejected: composed together: the checker's verdict at the members' heads (8a80e8e86858): unit: git.novox.be/n…
mesh/check failed (4): unit: git.novox.be/novox/m…; and 3 more | 62fe1cc36d30 on f2b332de73d6 | judged by the candidate's mesh-build-go at e8757a23634d
Quick fix: add the one require to mesh-build-go's go.mod (a minute). Thorough fix, chosen because the kind was met twice (a
local go.work resolving what go.mod never names) and hides itself from every local build: checks/requiredmodules reads every
go.mod and Go source (tests and tagged files included), finds the owning module by the longest module line among the
repositories beside, and fails in go test and the repository check before any push. The scan over the seven group heads
found this one instance.
Claude-Session: https://claude.ai/code/session_01NcbgfFSnn24M3GcqttoU7i
mesh/check failed (3): ace: the check of novox/mesh-catalog#612 failed: the unit level: the candidate's mes…; and 2 more | 7bf15a3dc4db on f2b332de73d6
mesh/check being carried to c8bcb2cdf567 on f2b332de73d6 | from 980ade933853, hq 885099eebbbd, mesh-delivery 2d201f781ced, mesh-sdk c1b50e99a639
The operator's answer: reading mesh-delivery-checker.inputs broke ADR 0596 §2. The manifest drops that read and invokes
seat:mesh-delivery-checker.input; the run step asks the verb through the runtime (stdio.Ask) and takes only the bytes
the ask names (delivery.RunInput.Read).
Claude-Session: https://claude.ai/code/session_01NcbgfFSnn24M3GcqttoU7i
"cmd/a/main.go":`package main; const reach = "MESH_REACH"; var _ = os.Getenv(reach); var _ = os.Getenv("MESH_SET")`,
"cmd/a/other.go":`package main; var _, _ = os.LookupEnv("MESH_LITERAL"); var _ = os.Getenv("MESH_MODULE")`,
"cmd/a/write.go":`package main; var _ = []string{"MESH_WRITTEN_FOR_A_CHILD=1"}`,
"cmd/a/a_test.go":`package main; var _ = os.Getenv("MESH_IN_A_TEST")`,
"client.ts":`const a = process.env.MESH_TS ?? "d"; const b = env["MESH_BRACKET"]; const c = env.MESH_VARIABLE;`,
"cmd/a/main.go":`package main; const reach = "MESH_REACH"; var _ = os.Getenv(reach); var _ = os.Getenv("MESH_SET")`,
"cmd/a/other.go":`package main; var _, _ = os.LookupEnv("MESH_LITERAL"); var _ = os.Getenv("MESH_MODULE")`,
"cmd/a/write.go":`package main; var _ = []string{"MESH_WRITTEN_FOR_A_CHILD=1"}`,
"cmd/a/a_test.go":`package main; var _ = os.Getenv("MESH_IN_A_TEST")`,
// 47 H2 §781, H-A-370: a declared setting's word, which the node-engine composes; one undeclared is set by nothing.
"client.ts":`const a = process.env.MESH_TS ?? "d"; const b = env["MESH_BRACKET"]; const c = env.MESH_VARIABLE; const d = env.MESH_CONFIG_ASK_TAKEN_BOUND; const e = env.MESH_CONFIG_UNDECLARED;`,
// passing line in its place). A build module with no test step (mesh-build-image builds images and runs no tests) is
// exempt from that, and said so. Every build module, test step or not, must also hold a test that calls
// conformance.LeftRunning with its own start sweep, so a step a previous process left running is ended at its start
// (issue 1890).
// (issue 1890), and one that calls conformance.Passing, so a step whose network did not answer is said as a transport fault
// (issue 1748), and one that calls conformance.StepEnds, so its steps can say a pipeline step's end (design 47 G §24). Its manifest asks the capability seat-settle-delay (ADR 0664 §5).
packagefailurekit
import(
@@ -13,6 +14,7 @@ import (
"io/fs"
"os"
"path/filepath"
"slices"
"sort"
"strings"
)
@@ -27,9 +29,15 @@ type manifest struct {
Claims[]struct{
Namestring`json:"name"`
}`json:"claims"`
Tools[]json.RawMessage`json:"tools"`
Tools[]json.RawMessage`json:"tools"`
Capabilities[]string`json:"capabilities"`
}
// settleDelay is the capability a tool runner states when it can settle a seat's ask with a delayed negative acknowledgement
// (mesh-sdk CapSeatSettleDelay): a build module gives its ask again after a delay by it, so the delivery plan places a tool
// runner that states it before the module (ADR 0664 §1, §5).
constsettleDelay="seat-settle-delay"
// Judge reads every modules/<m>/module.json of the catalogue at root and judges each module that claims a node-build-*
refused=append(refused,fmt.Sprintf("it claims %s, and no test of it calls conformance.RunsCounted "+
"(mesh-delivery-sdk steprun/conformance): run the kit's case that its puller counts the runs that started, not the bus's deliveries (A-489a)",seat))
}
if!ends{
refused=append(refused,fmt.Sprintf("it claims %s, and no test of it calls conformance.StepEnds "+
"(mesh-delivery-sdk steprun/conformance): run the kit's step-ends case with the module's own steps (design 47 G §24)",seat))
}
// Every build module meets the network (a fetch, a download, an install, a publish), so every one runs the passing case: its
// steps say a network that did not answer as a transport fault the bus gives the ask again for (ADR 0664, issue 1748).
says,err:=runsTheKit(dir,"conformance.Passing(")
iferr!=nil{
returnnil,err
}
if!says{
refused=append(refused,fmt.Sprintf("it claims %s, and no test of it calls conformance.Passing "+
"(mesh-build-go buildstep/conformance): run the kit's passing case with the module's own path from a failed run to its step's error (issue 1748)",seat))
}
// Every build module gives a step that met a transport fault again by the bus (ADR 0664 §5): it asks the tool runner's
// capability in its manifest, so the delivery plan places a tool runner that states it first.
if!slices.Contains(m.Capabilities,settleDelay){
refused=append(refused,fmt.Sprintf("it claims %s, and its manifest's capabilities do not ask %s: a step given again by the bus needs "+
"a tool runner that can settle with a delay (ADR 0664 §5, issue 1748)",seat,settleDelay))
"modules/i/module.json":`{"claims":[{"name":"node-build-i"}],"capabilities":["other"],"tools":["build_i_build"]}`,// the whole kit, and no seat-settle-delay
"modules/i/cmd/i/kit_test.go":imp+sweep,
"modules/j/module.json":`{"claims":[{"name":"node-build-j"}],"capabilities":["seat-settle-delay"],"tools":["build_j_build"]}`,// the rest of the kit, and no step-ends case
continue// the manifest validator names a malformed seat read
}
owners[owner]=a
for_,p:=rangeproblems{
out=append(out,owner+": "+p)
targets,state=holders[seat],local
}elseifowner,local,ok:=strings.Cut(r,".");ok{
targets,state=[]string{owner},local
}
for_,owner:=rangetargets{
dir,inCatalogue:=dirs[owner]
if!inCatalogue||owner==reader{
continue
}
a,seen:=owners[owner]
if!seen{
varproblems[]string
a,problems,err=AnnouncedBy(dir)
iferr!=nil{
returnnil,err
}
owners[owner]=a
for_,p:=rangeproblems{
out=append(out,owner+": "+p)
}
}
ifa==nil{
continue// announces nothing from code: read as before
}
readable,keeps:=a[state]
switch{
case!keeps:
out=append(out,fmt.Sprintf("%s reads %s.%s, which %s's code does not announce: the read is refused on the mesh",reader,owner,state,owner))
case!readable:
out=append(out,fmt.Sprintf("%s reads %s.%s, which %s's code announces not readable: the read is refused on the mesh; announce it readable",reader,owner,state,owner))
}
}
ifa==nil{
continue// announces nothing from code: read as before
}
readable,keeps:=a[state]
switch{
case!keeps:
out=append(out,fmt.Sprintf("%s reads %s.%s, which %s's code does not announce: the read is refused on the mesh",reader,owner,state,owner))
case!readable:
out=append(out,fmt.Sprintf("%s reads %s.%s, which %s's code announces not readable: the read is refused on the mesh; announce it readable",reader,owner,state,owner))
out=append(out,fmt.Sprintf("%s: uses %s.%s, which is the build runner's, no longer mesh-sdk's buildstep; import git.novox.be/novox/mesh-catalog/modules/mesh-build-go/buildstep (novox/hq ADR 0593 §1)",rel,buildstep,sel.Sel.Name))
out=append(out,fmt.Sprintf("%s: uses %s.%s, which is the build runner's, no longer mesh-sdk's buildstep; import git.novox.be/novox/mesh-delivery/modules/mesh-delivery-sdk/steprun (novox/hq ADR 0593 §1)",rel,buildstep,sel.Sel.Name))
{"a removed package","modules/a/x.go",head+"\t\"git.novox.be/novox/mesh-sdk/go/compose\"\n)\n\nvar _ = compose.Resolve\n","imports git.novox.be/novox/mesh-sdk/go/compose, which left mesh-sdk; it is the node-engine"},
{"under a removed package","modules/a/x_test.go",head+"\t\"git.novox.be/novox/mesh-sdk/go/compose/frombus\"\n)\n","the node-engine"},
{"the runner through mesh-sdk's buildstep","modules/a/x.go",head+"\t\"git.novox.be/novox/mesh-sdk/go/buildstep\"\n)\n\nvar _ = buildstep.RunAsk\n","uses buildstep.RunAsk, which is the build runner's"},
{"the runner through an alias","modules/a/x.go",head+"\tsdkstep \"git.novox.be/novox/mesh-sdk/go/buildstep\"\n)\n\nvar _ = sdkstep.Puller{}\n","uses sdkstep.Puller"},
{"the contract of buildstep","modules/a/x.go",head+"\t\"git.novox.be/novox/mesh-sdk/go/buildstep\"\n)\n\nvar _ = buildstep.NewID\nvar _ buildstep.Failure\n",""},
{"the runner's own package","modules/a/x.go",head+"\t\"git.novox.be/novox/mesh-catalog/modules/mesh-build-go/buildstep\"\n)\n\nvar _ = buildstep.RunAsk\n",""},
{"the runner's own package","modules/a/x.go",head+"\t\"git.novox.be/novox/mesh-delivery/modules/mesh-delivery-sdk/steprun\"\n)\n\nvar _ = steprun.RunAsk\n",""},
{"a copy in testdata","modules/a/testdata/x.go",head+"\t\"git.novox.be/novox/mesh-sdk/go/compose\"\n)\n",""},
{"a required module in a block",map[string]string{"a/go.mod":aMod+"\nrequire (\n\tgit.novox.be/novox/r/b/sub v0.0.0 // indirect\n)\n","a/x.go":src("git.novox.be/novox/r/b/sub/pkg"),"b/go.mod":bMod},""},
{"a module not required",map[string]string{"a/go.mod":aMod,"a/x.go":src("git.novox.be/novox/r/b/sub/pkg"),"b/go.mod":bMod},"x.go imports git.novox.be/novox/r/b/sub/pkg"},
{"a test file",map[string]string{"a/go.mod":aMod,"a/x_test.go":src("git.novox.be/novox/r/b/sub/pkg"),"b/go.mod":bMod},"x_test.go imports"},
{"a module the longest owner is not",map[string]string{"a/go.mod":aMod+"\nrequire git.novox.be/novox/r/b v0.0.0\n","a/x.go":src("git.novox.be/novox/r/b/sub/pkg"),"b/go.mod":"module git.novox.be/novox/r/b\n","b/sub/go.mod":bMod},"does not require"},
@@ -119,3 +119,16 @@ removed for `retention` on its seat (`release`, `remove`). It reads a reference
and `<seat>/<module>/<artifact>@<digest>`, as one artifact, and says artifacts-gone in the form the build named it. A
call a seat refuses stops the sweep and writes the finding `sweep-refused` in the module's findings; the next sweep no
seat refuses clears it (`sweepseats.go`).
### One pending pass
A finished build or a changed want marks the one pending pass (design 47 H, H-A-250); a pass running when one is marked
runs once more when it ends, however many events marked it. The builder's builds are a followed view: the watch's first view
at start, then its updates, kept in memory (H-A-255), each build's end taken from the builder's log's build-ended entries the
pass reads once, and only for a build the log lacks from its key's history (H-A-258); a watch asked again after an error replaces the view whole with its handover (H-A-257). A pass reads nothing
of them on the bus, judges nothing before the first view is in, and sweeps and observes S25
from that one judgement. A list the registry
seats cannot answer ends the pass with the finding `artifact-list-unread`, written once; nothing asks the list again
until the next event, and a pass that reads it clears the finding. A pass that cannot judge the builds (the view not yet taken in, or a
done build's end unread) ends the same way with the finding `artifact-builds-unread` (H-A-260) (`holder.go`,
@@ -28,8 +28,7 @@ And, for where the agent runs as an account of its own (hq ADR 0266):
rule and can read no secret the mesh placed for another account. A way found is this module's condition,
and the controller's self-check says it urgently.
-`/usr/local/bin/claude-agent`, root's, `0755`: starts the agent as that account — `claude` itself when run
as it, otherwise `sudo -iu <account> claude` through the operator's own sudo, carrying the guard's
override only when it is set in the operator's shell.
as it, otherwise `sudo -iu <account> claude` through the operator's own sudo.
-`/etc/claude-code/agent-account`, root's, `0644`: the name of that account, the machine fact
`agent-account`, for the shell to read — the controller fills no machine fact into shell code.
- **`claude` in the operator's interactive zsh is `claude-agent`** where the machine names an agent account:
@@ -48,7 +47,7 @@ whenever the node's tool runtime collects the module's tools:
| `managed-mcp.json` | the tool servers every session loads: the mesh's console as `mesh` (with a `timeout` of 660000 ms, mesh_wait's longest wait and a minute, so the agent's one-minute default does not cut a wait, issue 1871), and the servers set in this module's `mcp_servers` setting. **Exclusive**: a server not listed here does not load — not one added with `claude mcp add`, not a project's `.mcp.json`, not a plugin's |
| `managed-settings.json` | the keys set in this module's `managed_settings` setting — at the controller's terminal alone (hq issue 340) — under the mesh's own keys: the repositories' attribution convention, the claude.ai connectors kept beside the managed servers, the key-helper while the node holds an API-key licence, git's credential file while the node holds the forge's read credential (below), and the two that name the `nox-mesh` marketplace and enable its plugin, and `CLAUDE_CODE_MCP_TOOL_IDLE_TIMEOUT` (660000) in `env` beside the operator's own, the mesh's key winning |
| `CLAUDE.md` | how a session on this mesh works, this node's name and role, the conventions, the **"instead of" table** — then the instruction sections registered for every node and for this one |
| `marketplace/` | the `nox-mesh` plugin (hq ADR 0216): the skills, subagents, commands and output styles registered for every node and for this one, offered in a session as `nox-mesh:<name>` — and the **guard on the agent's shell** (`guard/`), the mesh's own hook and the plugin's only one (hq issue 340). Replaced whole, staged beside and swapped in |
| `marketplace/` | the `nox-mesh` plugin (hq ADR 0216): the skills, subagents, commands and output styles registered for every node and for this one, offered in a session as `nox-mesh:<name>` — and this module's binary (`guard/guard`, its path kept so the home-fs helper's path does not move) run by the mesh's own SessionStart and Stop hooks, the plugin's only hooks (hq issue 340, ADR 0489). No PreToolUse hook: the guard on the agent's shell is removed (the operator, 2026-11-20). Replaced whole, staged beside and swapped in |
Under the operator's home: `~/.claude/.credentials.json`, only when the licence manager hands this node a
subscription token; and what is registered at the **home** scope for this node — a skill, subagent,
@@ -78,12 +77,12 @@ must see, a node that joins later included — kept, so it carries no secret eit
| what | how |
|---|---|
| what this node holds | the module's `holdings` state, one key for this node — the account, the kind, fingerprints and expiries, never a token — written at start and whenever the credentials file changes (hq ADR 0206) |
| what this node holds | the module's `holdings` state, one key for this node — the account, the kind, fingerprints and expiries, never a token — written at start and whenever the credentials file changes (hq ADR 0206), and said whole as the event `holdings-changed` at start, on every change and again when the licence manager says `holdings.asked` (design 39 bus-and-switch §23) |
| a person ran `/login` here | the credentials file gains a refresh token this module never writes; its next report shows it, and the licence manager asks `claude_code_grant` for it, giving its key — the one time a refresh token leaves the node, for the manager to adopt by refreshing it |
| what this node should hold | the licence manager's `bindings` state, this node's key; on a newer generation this module asks `anthropic-licence-manager.current` for its token, sealed to the key it sends, and writes it access-token-only — so the agent here never refreshes. A node that was off reads its key when it is back |
| what this node should hold | the licence manager's `binding.changed` naming this node, and at start (or a reconnect) its first view, `anthropic-licence-manager.bindings` with `consumer` this node; an event for another node is ignored. On a newer generation, at start, and on the vault's `secret-changed` for the licence it holds, this module asks `anthropic-licence-manager.current` which licence, and the vault's `get` for that licence's access token (`licence.<key>.access`), sealed to its own key, and writes it access-token-only — so the agent here never refreshes, and never receives a refresh token. A home holding no token is fetched again at the same generation (design 39 bus-and-switch §19, §22) |
| an MCP server registered through this module before registering was refused (hq issue 340) | a key in the module's `servers` state — `all.<server>` for every node, `<node>.<server>` for one; every node watches it and renders what applies to it, a node's own entry over the one for every node. A node that joins later, or was off, reads the whole current set at start; unregistering is a delete. Nothing adds a key any more: a new tool server is the operator's, in the `mcp_servers` setting at the controller's terminal |
| a proposed instruction section (hq ADR 0274) | a key in the module's `proposals` state per proposal, the ask published on `operator-channel` under this module's name, and the router's record of it read under that name alone; what a node approved is its own file, never the bus |
| a session's question to the operator (hq ADR 0332) | a key `q-…` in the module's `questions` state per question, the ask published on `operator-channel` under this module's name, and the router's record of it read under that name alone; the router's word that it ended without a choice is kept on the question; two slot keys `slot-1`, `slot-2` name the questions open |
| a session's question to the operator (hq ADR 0332) | a key `q-…` in the module's `questions` state per question, the ask published on `operator-channel` under this module's name, and the router's record of it read under that name alone; the router's word that it ended without a choice is kept on the question |
| a project's memory (hq ADR 0343) | a key in the module's `memory` state per file — `<kind>.<project and path, base64url>`, `operator` for the operator's own homes and `agent` for an agent account's — holding the text, its digest, the digest of the version it replaced, the node and when it changed; a removal is a value marked removed, never a deleted key. Written at the revision read, so of two nodes writing one file only one wins. Every node of the same kind watches it. No event and no tool answer carries memory text |
| the agent's configuration (hq ADR 0216) | a key in the module's `config` state per registration — `mesh.<kind>.<name>` for every node, `node.<node>.<kind>.<name>` for one, `home.<node>.<kind>.<name>` for one account's own directory — the item and its files in one value, at most 256 KiB. Every node watches it and renders what applies to it, a node item over a mesh item of the same kind and name. A hook, settings, or an item that would run something by itself is never taken, whenever it was put there (hq issue 340) |
@@ -126,32 +125,16 @@ The agent kept reaching for `ssh <machine> journalctl` while the service manager
answered this is not placed) — at start, every ten minutes and on `claude_code_render`, keeps it in its
state (`mesh-tools.json`), and renders it into `CLAUDE.md` as one row per seat or module: `<node>/<seat>.` and
each verb with the commands it replaces. Generated, never written by hand: a verb that gains `replaces` is in
the next render. Ordered by what the guard here refused most, then the machines' seats, the modules, the
the next render. Ordered by the machines' seats, the modules, the
mesh's seats; at most sixteen rows, the rest one `mesh_search` away.
- **The guard.** A PreToolUse hook on every tool, judging`Bash`, `Edit`, `Write`, `MultiEdit` and `NotebookEdit` and letting the rest through (but see the account rule below): this module's own
binary (`claude-code guard`), copied root's into the plugin with what it judges with (`guard.json`: the
machines by name, domain and address, and the replaced commands), run by its path under `/etc/claude-code` so
the session cannot change it. It refuses `ssh`, `scp`, `sftp`, `rsync`, `mosh` and `autossh` to a mesh machine
(any `*.internal` name, a machine's name or a name under its domains, one of its addresses, as `ssh -G` reads
the destination; a jump through one too), writing `/etc/hosts` or `/etc/resolv.conf`, and `HOSTALIASES` —
naming the verb that does the job on that machine when one says it replaces the command, and otherwise that a
missing tool is created in the module that owns it, never worked around. **Stated, not silent:** an ssh login
as `git` is the forge's account, which runs nothing but git, and passes; a git remote is never an ssh command
line anyway.
- **The operator's override**: `MESH_GUARD_OVERRIDE=<why>`, exported in the operator's own shell before the
session starts. It is read from the session's environment as the kernel kept it at exec
(`/proc/<pid>/environ` of the agent's process), so nothing a session does — a command's `export`, a
settings `env` key — can set it, and a command naming it is refused outright. Every override and every
refusal is a line in the module's `guard.log`; an override that cannot be recorded is not honoured.
`claude_code_guard` shows the rules, the data and the record.
It is a guard against the habit, not a sandbox: a command written to hide what it runs can hide it, and the
record is how a habit that found a way round is seen.
The rule under the table (never`ssh` to a mesh machine, never edit `/etc/hosts` or set `HOSTALIASES` for a
mesh name) is the session's own: nothing on the agent's shell enforces it since the guard was removed (the
operator, 2026-11-20).
## Tools
`claude_code_status`, `claude_code_render`, `claude_code_guard`, `claude_code_pull`, `claude_code_grant` (for the licence
manager), `claude_code_mcp_list`, `claude_code_mcp_unregister` (this node by default; `nodes: "all"` or a list for
`claude_code_status`, `claude_code_render`, `claude_code_pull` (asks the vault for this node's
licence token now), `claude_code_grant` (for the licence manager), `claude_code_mcp_list`, `claude_code_mcp_unregister` (this node by default; `nodes: "all"` or a list for
more).
**What a session runs or obeys is the operator's, at the controller's terminal** (hq issue 340). A hook, a
@@ -233,11 +216,8 @@ key says waiting, chosen (the option, who answered, on which channel, when), end
refused, cancelled, replaced), or not proven: the answer is read from the router's own record and held to the
question as `questions` keeps it, so a question rewritten after it was asked, a warrant for another ask or a choice
after expiry is never read as a choice. One call waits at most 60 s; the tool asks a session to read a question at
most once a minute, and nothing enforces it. At most two questions wait at once, mesh-wide: each holds one of two
slot keys in `questions`, taken by create or compare-and-set, so two nodes asking at once cannot both take the last,
and a slot that cannot be read refuses the question. Questions alone never fill the router's three open asks, so one
is left for a proposed instruction section while two are open. A question the router holds no record of, its refusal
not heard here, keeps its slot and reads as waiting until it expires: the router may still take it from its queue,
most once a minute, and nothing enforces it. No question is refused for how many wait: the router offers an asker at most three asks at once, holds the oldest back behind newer ones and offers it again when one ends (hq ADR 0666 §12). A question the router holds no record of, its refusal
not heard here, reads as waiting until it expires: the router may still take it from its queue,
and no cancel can overtake it there (the queue offers work again after an error, and a cancel of an ask not yet
recorded is dropped); once expired, the router refuses it if taken late. A refused question keeps none of
its words. An answer is the operator's word to the session and authorises nothing; a question never defaults.
@@ -342,14 +322,6 @@ applied only while the agent's account holds it — and what this module placed
goes: the access-token-only credentials file (one holding a refresh token is a login the person made, and
stays, said), the home-scope items unless changed by hand, the key kept in the state.
**The guard there refuses every tool call of a session that does not run as the agent's account** —
shell, file edits, reads and the mesh's tools alike — and no override lifts it: the operator's account may become root, and an
agent started in it by habit must do nothing. Its record goes to the machine's journal
(`journalctl -t mesh-claude-code-guard`), which the session's account may add to and not change;
`claude_code_guard` reads it there, keeping only the lines the journal says the guard's own binary wrote. That
is not proof: the account can run the guard's binary itself and add lines in the guard's words. What it
cannot do is remove one.
**An API key is taken only from the agent's own home.** `claude_code_add_api_key` reads a file and removes it,
and the runtime runs as the operator: a path anywhere would hand whoever calls it the runtime's credential or
the operator's keys. So the file must lie under the home of the account the agent runs as, reached through no
{"a login handed, met by its refresh: same lineage, later access expiry (H-A-139)",login,true,9000,now+3*7_200_000,func(pPaths,askAsk,wWriteManaged)(string,error){
{"a login handed, met by the same lineage with an equal access expiry (H-A-139)",login,true,9000,now+7_200_000,func(pPaths,askAsk,wWriteManaged)(string,error){
{"not handed, the same account, a lease that succeeds it: the latest login wins (H-A-140)",login,false,9000,now+3*7_200_000,func(pPaths,askAsk,wWriteManaged)(string,error){
// The hand-act seat's warranted and hand-act, the seat mesh-catalog's seat list (ADR 0635 §1), and the owners' read tools of the three followed states; no verb of
// the controller, and no read of the indexer's or the assigner's state (ADR 0615 §1): their views are consumed.
// the controller, and no read of the indexer's or the assigner's state (ADR 0615 §1): their views are consumed. The
// licence binding is the seat's bindings and the manager's binding.changed, its token the vault's get and secret-changed (design 39 bus-and-switch §19, §22, B9); a
// notice goes out on the operator channel's notify (design 47 G GR8).
b[ModulesRead][key],_=json.Marshal(catalog.ModulesByModuleValue{V:1,None:"nothing is built of it",Manifest:digest,Repository:"novox/mesh-catalog",Commit:"c"})
t.Fatalf("from the views: %+v, %v\nfrom the state: %+v",got,err,want)
}
// A reader not yet answered says so, and reads nothing.
if_,err:=ReadMeshToolsFrom(viewStates{newMeshViews()},someSeats(),resolving);err==nil||!strings.Contains(err.Error(),"no first view"){
// A catalog half with no first view says so, and reads no module as absent.
if_,err:=ReadMeshToolsFrom(bus,fakeCatalog{notYet:errors.New("the catalog half has no first view of main yet")},someSeats());err==nil||!strings.Contains(err.Error(),"no first view"){
t.Fatalf("an unanswered catalog half: %v",err)
}
if_,err:=ReadMeshToolsFrom(viewStates{newMeshViews()},aCatalog(),someSeats());err==nil||!strings.Contains(err.Error(),"no first view"){
@@ -9,5 +9,5 @@ func instructionsText(node, role string) string {
node+
"`\n- **Role:** "+
role+
"\n- The other nodes, their roles and what runs where: ask the seat `mesh-module-assigner` (`mesh-module-assigner.assignments`),\n or `mesh_overview` and `mesh_machine`. Nothing here lists them, because a copy drifts.\n\n## How a session on this mesh works\n\nThe console is the only way to the mesh: the MCP server named `mesh`. It offers five tools, and\neverything else is an address you find and call through them:\n\n- `mesh_search` — words in, matching addresses out. `mesh_describe` — one address's arguments.\n- `mesh_call` — call an address. A seat the mesh holds once is `<seat>.<verb>` (the mesh's own seats\n include `mesh-module-assigner`, `mesh-delivery`, `mesh-config` and `mesh-conditions`);\n a module on a machine is `<node>/<module>.<tool>`.\n- `mesh_overview` and `mesh_machine` — the mesh's seats and machines, and what one machine runs.\n\n- **Symptom first.** For an error, a failing service or anything unexpected, search the record with the\n literal text before forming a hypothesis: the records module's `records_search`, then\n `records_read`.\n- **Ask the mesh before changing it**, and change it through the seats' verbs or the catalog.\n- **A licence** through the `anthropic-licence-manager` seat's verbs. Never edit the agent's credentials\n file by hand, never print or ask for a token.\n\n## Hard rules\n\n- A file the mesh manages is changed through the verb or the catalog that owns it, never on disk. If\n unsure, `<node>/node-engine.composition` for the module says what the mesh writes there.\n- Never write to a store's database by hand; schema changes are numbered migrations.\n- Never push to a main branch: a branch, a pull request, and a human approval for every merge.\n- The mesh creates no symlinks, and nobody else does either.\n- A package is declared in a module, never installed by hand.\n\n## Conventions\n\n- Commit messages are concise, in the imperative, about why.\n- Test before pushing: nodes update unattended.\n- The playbooks in the record say how research, decisions, designs, issues and hand-offs are done.\n"
"\n- The other nodes, their roles and what runs where: ask the seat `mesh-module-assigner` (`mesh-module-assigner.assignments`),\n or `mesh_overview` and `mesh_machine`. Nothing here lists them, because a copy drifts.\n\n## How a session on this mesh works\n\nThe console is the only way to the mesh: the MCP server named `mesh`. It offers five tools, and\neverything else is an address you find and call through them:\n\n- `mesh_search` — words in, matching addresses out. `mesh_describe` — one address's arguments.\n- `mesh_call` — call an address. A seat the mesh holds once is `<seat>.<verb>` (the mesh's own seats\n include `mesh-module-assigner`, `mesh-delivery`, `mesh-config` and `mesh-conditions`);\n a module on a machine is `<node>/<module>.<tool>`.\n- `mesh_overview` and `mesh_machine` — the mesh's seats and machines, and what one machine runs.\n\n- **Symptom first.** For an error, a failing service or anything unexpected, search the record with the\n literal text before forming a hypothesis: the records module's `records_search`, then\n `records_read`.\n- **Ask the mesh before changing it**, and change it through the seats' verbs or the catalog.\n- **A licence** through the `anthropic-licence-manager` seat's verbs. Never edit the agent's credentials\n file by hand, never print or ask for a token.\n\n## Hard rules\n\n- A file the mesh manages is changed through the verb or the catalog that owns it, never on disk. If\n unsure, `<node>/node-engine.composition` for the module says what the mesh writes there.\n- Never write to a store's database by hand; schema changes are numbered migrations.\n- Never push to a main branch: a branch and a pull request for every change.\n- The mesh creates no symlinks, and nobody else does either.\n- A package is declared in a module, never installed by hand.\n\n## Conventions\n\n- Commit messages are concise, in the imperative, about why.\n- Test before pushing: nodes update unattended.\n- The playbooks in the record say how research, decisions, designs, issues and hand-offs are done.\n"
Description:"Claude Code on this machine as the mesh configured it: the licence it holds and when its token expires, what it reports holding, the managed files, the MCP servers registered here. Fingerprints only, never a token.",
Description:"Write Claude Code's managed directory now, from the mesh's facts, this module's settings and the servers registered here — after reading the bus again for what each tool replaces and which machines the mesh has (the instructions' table and the shell's guard).",
Description:"Write Claude Code's managed directory now, from the mesh's facts, this module's settings and the servers registered here — after reading the bus again for what each tool replaces (the instructions' table).",
Description:"The guard on the agent's shell on this machine (novox/hq ADR 0245): what it refuses — ssh to a mesh machine, writing /etc/hosts or /etc/resolv.conf, HOSTALIASES — the machines and the replaced commands it judges with, the \"instead of\" table rendered into the agent's instructions, and its record of what it refused and what the operator's override let through, newest last.",
Input:map[string]any{"last":map[string]any{"type":"integer","description":"how many lines of the record (default 50)"}},
Run:func(amap[string]any)(any,error){
last:=50
ifn,ok:=a["last"].(float64);ok&&n>0{
last=int(n)
}
returnGuardStatus(p,last),nil
}},
{Name:"claude_code_pull",
Description:"Ask the licence manager for this node's current token now and apply it, rather than waiting for its binding to change.",
Description:"Ask the licence manager which licence this node holds and the vault for its access token now, and apply it, rather than waiting for its binding or the token to change.",
Description:"For the licence manager (ADR 0206): the full grant in this node's credentials file — a login made here — sealed to the public key given, with the account it belongs to. Nothing when no login is waiting. Never answers a token in the clear.",
// ReadMeshToolsFrom reads the bus now: the seats' verbs and the placed modules' own tools. A part that cannot be read is left out and said in the error; what was read is kept.
log.Printf("login superseded: a later login to the account of licence %s was adopted by the manager (generation %d)",c.Licence,c.Generation)
}
// A login handed to the manager is replaced only by a grant of a newer lineage (its refresh token's
// expiry later than the login's, or the same lineage with a later access expiry, as the manager's refresh yields it): a lease older than the login is never
// written, so the manager's failed refresh is a finding on the manager, never a lost login here
"CLAUDE.md":"# This machine is a node of a Novox mesh\n\nWritten by the mesh's `claude-code` module. Edit the module's settings or the catalog, never this file:\nit is rewritten whenever the module renders.\n\n## Who this node is\n\n- **Node:** `workstation`\n- **Role:** the laptop\n- The other nodes, their roles and what runs where: ask the seat `mesh-module-assigner` (`mesh-module-assigner.assignments`),\n or `mesh_overview` and `mesh_machine`. Nothing here lists them, because a copy drifts.\n\n## How a session on this mesh works\n\nThe console is the only way to the mesh: the MCP server named `mesh`. It offers five tools, and\neverything else is an address you find and call through them:\n\n- `mesh_search` — words in, matching addresses out. `mesh_describe` — one address's arguments.\n- `mesh_call` — call an address. A seat the mesh holds once is `<seat>.<verb>` (the mesh's own seats\n include `mesh-module-assigner`, `mesh-delivery`, `mesh-config` and `mesh-conditions`);\n a module on a machine is `<node>/<module>.<tool>`.\n- `mesh_overview` and `mesh_machine` — the mesh's seats and machines, and what one machine runs.\n\n- **Symptom first.** For an error, a failing service or anything unexpected, search the record with the\n literal text before forming a hypothesis: the records module's `records_search`, then\n `records_read`.\n- **Ask the mesh before changing it**, and change it through the seats' verbs or the catalog.\n- **A licence** through the `anthropic-licence-manager` seat's verbs. Never edit the agent's credentials\n file by hand, never print or ask for a token.\n\n## Hard rules\n\n- A file the mesh manages is changed through the verb or the catalog that owns it, never on disk. If\n unsure, `<node>/node-engine.composition` for the module says what the mesh writes there.\n- Never write to a store's database by hand; schema changes are numbered migrations.\n- Never push to a main branch: a branch, a pull request, and a human approval for every merge.\n- The mesh creates no symlinks, and nobody else does either.\n- A package is declared in a module, never installed by hand.\n\n## Conventions\n\n- Commit messages are concise, in the imperative, about why.\n- Test before pushing: nodes update unattended.\n- The playbooks in the record say how research, decisions, designs, issues and hand-offs are done.\n"
"CLAUDE.md":"# This machine is a node of a Novox mesh\n\nWritten by the mesh's `claude-code` module. Edit the module's settings or the catalog, never this file:\nit is rewritten whenever the module renders.\n\n## Who this node is\n\n- **Node:** `workstation`\n- **Role:** the laptop\n- The other nodes, their roles and what runs where: ask the seat `mesh-module-assigner` (`mesh-module-assigner.assignments`),\n or `mesh_overview` and `mesh_machine`. Nothing here lists them, because a copy drifts.\n\n## How a session on this mesh works\n\nThe console is the only way to the mesh: the MCP server named `mesh`. It offers five tools, and\neverything else is an address you find and call through them:\n\n- `mesh_search` — words in, matching addresses out. `mesh_describe` — one address's arguments.\n- `mesh_call` — call an address. A seat the mesh holds once is `<seat>.<verb>` (the mesh's own seats\n include `mesh-module-assigner`, `mesh-delivery`, `mesh-config` and `mesh-conditions`);\n a module on a machine is `<node>/<module>.<tool>`.\n- `mesh_overview` and `mesh_machine` — the mesh's seats and machines, and what one machine runs.\n\n- **Symptom first.** For an error, a failing service or anything unexpected, search the record with the\n literal text before forming a hypothesis: the records module's `records_search`, then\n `records_read`.\n- **Ask the mesh before changing it**, and change it through the seats' verbs or the catalog.\n- **A licence** through the `anthropic-licence-manager` seat's verbs. Never edit the agent's credentials\n file by hand, never print or ask for a token.\n\n## Hard rules\n\n- A file the mesh manages is changed through the verb or the catalog that owns it, never on disk. If\n unsure, `<node>/node-engine.composition` for the module says what the mesh writes there.\n- Never write to a store's database by hand; schema changes are numbered migrations.\n- Never push to a main branch: a branch and a pull request for every change.\n- The mesh creates no symlinks, and nobody else does either.\n- A package is declared in a module, never installed by hand.\n\n## Conventions\n\n- Commit messages are concise, in the imperative, about why.\n- Test before pushing: nodes update unattended.\n- The playbooks in the record say how research, decisions, designs, issues and hand-offs are done.\n"
"CLAUDE.md":"# This machine is a node of a Novox mesh\n\nWritten by the mesh's `claude-code` module. Edit the module's settings or the catalog, never this file:\nit is rewritten whenever the module renders.\n\n## Who this node is\n\n- **Node:** `workstation`\n- **Role:** not stated — set it in this module's settings for the node\n- The other nodes, their roles and what runs where: ask the seat `mesh-module-assigner` (`mesh-module-assigner.assignments`),\n or `mesh_overview` and `mesh_machine`. Nothing here lists them, because a copy drifts.\n\n## How a session on this mesh works\n\nThe console is the only way to the mesh: the MCP server named `mesh`. It offers five tools, and\neverything else is an address you find and call through them:\n\n- `mesh_search` — words in, matching addresses out. `mesh_describe` — one address's arguments.\n- `mesh_call` — call an address. A seat the mesh holds once is `<seat>.<verb>` (the mesh's own seats\n include `mesh-module-assigner`, `mesh-delivery`, `mesh-config` and `mesh-conditions`);\n a module on a machine is `<node>/<module>.<tool>`.\n- `mesh_overview` and `mesh_machine` — the mesh's seats and machines, and what one machine runs.\n\n- **Symptom first.** For an error, a failing service or anything unexpected, search the record with the\n literal text before forming a hypothesis: the records module's `records_search`, then\n `records_read`.\n- **Ask the mesh before changing it**, and change it through the seats' verbs or the catalog.\n- **A licence** through the `anthropic-licence-manager` seat's verbs. Never edit the agent's credentials\n file by hand, never print or ask for a token.\n\n## Hard rules\n\n- A file the mesh manages is changed through the verb or the catalog that owns it, never on disk. If\n unsure, `<node>/node-engine.composition` for the module says what the mesh writes there.\n- Never write to a store's database by hand; schema changes are numbered migrations.\n- Never push to a main branch: a branch, a pull request, and a human approval for every merge.\n- The mesh creates no symlinks, and nobody else does either.\n- A package is declared in a module, never installed by hand.\n\n## Conventions\n\n- Commit messages are concise, in the imperative, about why.\n- Test before pushing: nodes update unattended.\n- The playbooks in the record say how research, decisions, designs, issues and hand-offs are done.\n"
"CLAUDE.md":"# This machine is a node of a Novox mesh\n\nWritten by the mesh's `claude-code` module. Edit the module's settings or the catalog, never this file:\nit is rewritten whenever the module renders.\n\n## Who this node is\n\n- **Node:** `workstation`\n- **Role:** not stated — set it in this module's settings for the node\n- The other nodes, their roles and what runs where: ask the seat `mesh-module-assigner` (`mesh-module-assigner.assignments`),\n or `mesh_overview` and `mesh_machine`. Nothing here lists them, because a copy drifts.\n\n## How a session on this mesh works\n\nThe console is the only way to the mesh: the MCP server named `mesh`. It offers five tools, and\neverything else is an address you find and call through them:\n\n- `mesh_search` — words in, matching addresses out. `mesh_describe` — one address's arguments.\n- `mesh_call` — call an address. A seat the mesh holds once is `<seat>.<verb>` (the mesh's own seats\n include `mesh-module-assigner`, `mesh-delivery`, `mesh-config` and `mesh-conditions`);\n a module on a machine is `<node>/<module>.<tool>`.\n- `mesh_overview` and `mesh_machine` — the mesh's seats and machines, and what one machine runs.\n\n- **Symptom first.** For an error, a failing service or anything unexpected, search the record with the\n literal text before forming a hypothesis: the records module's `records_search`, then\n `records_read`.\n- **Ask the mesh before changing it**, and change it through the seats' verbs or the catalog.\n- **A licence** through the `anthropic-licence-manager` seat's verbs. Never edit the agent's credentials\n file by hand, never print or ask for a token.\n\n## Hard rules\n\n- A file the mesh manages is changed through the verb or the catalog that owns it, never on disk. If\n unsure, `<node>/node-engine.composition` for the module says what the mesh writes there.\n- Never write to a store's database by hand; schema changes are numbered migrations.\n- Never push to a main branch: a branch and a pull request for every change.\n- The mesh creates no symlinks, and nobody else does either.\n- A package is declared in a module, never installed by hand.\n\n## Conventions\n\n- Commit messages are concise, in the imperative, about why.\n- Test before pushing: nodes update unattended.\n- The playbooks in the record say how research, decisions, designs, issues and hand-offs are done.\n"
Holds the `anthropic-licence-manager` seat: every Anthropic licence the mesh has, kept alive by one
rotation source, and handed to each consumer sealed (novox/hq ADR 0183, ADR 0206, design 39).
rotation source, its consumers moved off a licence near its limit by the manager itself (novox/hq ADR 0183,
ADR 0206, design 39 and its milestone bus-and-switch). Every grant and access token lives in the vault; this module
keeps none.
It is one manifest of four parts (design 39 parts §1): the manager bundle (`manager`, `cmd/manager`), which runs long
(the conditions judged at its start, the usage timer, the switch rule) and answers the seat's verbs and every act; the
tools bundle (`tools`, `cmd/tools`), which answers the reading tools and writes nothing, so they answer while the
manager fails; the settings file; and the module's account with its directory. Both bundles run as an account of its
own (`runs-as claude-licence-manager`, design 39 bus-and-switch C1), in a runtime of its own, never in the machine's
tool runner, so every state, log and secret below is issued to its own account. Its
account is issued what its manifest names: the asks under `invokes` (the vault, the operator channel, mesh-config and
a node's `claude_code_grant`), the events under `emits` and `consumes`.
## How a licence comes to exist
Nothing is configured. Every node running `claude-code`reports what it holds as that module's
`holdings` state — the account, fingerprints and expiries, never a token. This module reads every report
when it starts and watches them:
Nothing is configured. Every node running `claude-code`says what it holds as the event
`claude-code.holdings-changed` (the account, fingerprints and expiries, never a token), at its start, on every change,
and again when this module says `holdings.asked` (at every start, after its conditions are judged, and on `licences_resync`):
1. A report with a refresh token it does not hold is a **candidate**.
2. It asks that node's `claude_code_grant`, giving its public key, and receives the grant sealed to it.
3. **It refreshes it.** If the vendor exchanges the token, the grant is this module's — encrypted at rest
with the key the vault made for it — and from then on it is the only refresher. If not, the candidate
is recorded dead and nothing is adopted.
3. **It refreshes it**, under the account's lease. If the vendor exchanges the token, the licence is new, named by its
account's address and kept under that name's key (`@` as `_at_`, any other character outside letters, digits, `-`
and `.` as `_`); its grant and access token go to the vault (`licence-put`), and from then on this module is the
only refresher. If not, the candidate is recorded dead and nothing is adopted. An adoption whose key another licence
has is refused.
4. Several nodes logged in to one account: newest login first; the rest are never exchanged.
5. A node reporting that account and bound to nothing is bound to it.
5. The node it came from is bound to it; a node reporting that account and bound to nothing is bound to it.
Each node is then handed an access token only, so the agent there never refreshes, and a refresh token
appearing on a node later can only be a person's login — which wins if it refreshes.
From its next reading a new subscription has a room and is a target of the switch. An API key enters through `adopt`,
from a file on this module's node or sealed by a node's `claude_code_add_api_key`; it is never a target.
An API key enters through `adopt`, from a file on this module's node.
## What it keeps on the bus
## What each consumer holds
Each state asked for in `requires` by its template (ADR 0595), and the two logs declared in the `logs` section, the
form the runtime issues today (design 39 bus-and-switch C3); each read by `licences_kept`:
This module's `bindings` state: one key per consumer (a node's name) with the licence, its kind and a
generation that grows with every rotation and switch. `claude-code` watches its own key and, on a newer
generation, asks `current` with its public key.
| name | template | what |
|---|---|---|
| `licences` | state, history 1 | one key per licence: account, address, organisation, expiries, failures, adoption, the vault's generation, the latest reading; never a token |
| `bindings` | state, history 8 | one key per consumer (a node's name): licence, kind, generation, since, why and by |
| `leases` | state, a lifetime per key | `licence.<key>`, `account.<id>`: one refresh or adoption at a time, two minutes |
| `offered` | state, a lifetime per key | each grant a node offered and its outcome, 90 days |
| `usage` | log, 35 days, 256 MiB | every reading, one subject per licence key, with the vendor's answer whole |
| `audit` | log, 64 MiB | every act, one subject per kind (`switched`, `carried`, `settings` among them) |
| `carried` | state | `postgres`: the record of the carry-over from the store database, which ran once on 2026-11-12; nothing writes it again |
| `conditions` | state | each condition said raised, and since when |
The carry-over from the store database is done, and its code has left the module with its requirement of the
database and of the secret `grant-key` (design 39 parts §9, §10); at its start the manager waits for nothing and
clears a `carry-over-failed` it finds kept.
## Switching on usage
Every `usage_minutes` the manager reads each subscription's usage with its access token from the vault: the session,
weekly and sonnet windows, each with its reset time. A licence's room is 100 less its weekly percentage alone (the session
window takes no part in room, the switch, the warning or the conditions: the vendor retired the session limit), known only when it was read within `usage_stale_minutes` and its refresh is healthy. When a reading puts
a licence with consumers at or above `switch_at_pct` in its weekly window, every consumer bound to it moves, in one act, to the licence
with the most room whose weekly window is below `room_below_pct`: one `binding.changed` per consumer, one
`licence.switched`, and one `operator-channel.tell`. A consumer is never moved back by the rule.
Its conditions are said on the operator channel's condition subject, the router deciding where they go:
`licence.<key>.no-room` (urgent; no licence has room), `licence.<key>.near-limit` (at `warn_at_pct`, once per
crossing), `licence.<key>.usage-unread` and `settings.<setting>.invalid`.
verb("licences","Every licence the manager holds — account, kind, when its token and its refresh token expire, failures in a row, which consumers are bound to it. Never a token.",
returnappend([]stdio.Tool{
verb("licences","Every licence the manager holds — account, kind, when its token and its refresh token expire, failures in a row, which consumers are bound to it, its latest usage with its reset times, its room or why it is not known, whether it is where a full licence's consumers move next, when and from where it was adopted, and its open conditions. Never a token.",
verb("bindings","Which licence each consumer (a node's agent, by the node's name) is bound to, and the generation it was last given; with consumer, that consumer's alone.",
map[string]any{"consumer":str("a consumer; absent for every one")},
verb("usage","Usage readings, the latest first, for every licence or one.",
map[string]any{"licence":str("a licence; absent for all"),"limit":map[string]any{"type":"number","description":"how many readings (default 20)"}},
verb("usage","Usage readings, the latest first, for every licence or one: the flattened reading with its three reset times, none before since, at most limit (default 20, at most 2000), and with raw the vendor's answer whole.",
map[string]any{"licence":str("a licence; absent for all"),"since":str("a time, RFC 3339: no reading before it"),
"limit":map[string]any{"type":"number","description":"how many readings (default 20, at most 2000)"},
"raw":map[string]any{"type":"boolean","description":"also the vendor's answer whole"}},
verb("adopt","Adopt an API key — never as an argument: from a file on the manager's node (`file`), or sealed to the manager's `public-key` by a node's claude-code (`sealed`; its `claude_code_add_api_key` does this). Subscriptions are adopted from the nodes' logins by themselves.",
map[string]any{"name":str("the licence's name"),"file":str("a file on the manager's node holding the key"),
verb("current","For a consumer's agent module (ADR 0206): its token, sealed to the public key it sends, with the licence, kind and generation. Null when it is bound to nothing.",
map[string]any{"consumer":consumer,"public_key":str("the consumer's public key, PEM")},
verb("current","For a consumer's agent module: which licence it is bound to, the licence's key, kind and generation, its health (ok, failing or expiring) and when its access token expires; never the token, which the vault hands. Null when it is bound to nothing.",
// A licence's room (design 39 bus-and-switch §25, §48, C4): 100 less its weekly percentage in its latest reading, the
// session window taking no part (the vendor retired the session limit), known only for a subscription read successfully within usage_stale_minutes, whose refresh has not
// failed failures_to_notify times in a row and whose refresh token has not expired. An API key's room is never known,
// so it is never a target.
import(
"time"
."claude-licence-manager/internal/licences"
)
// Room is a licence's room at now, and whether it is known.
Some files were not shown because too many files have changed in this diff
Show More
Reference in New Issue
Block a user
Blocking a user prevents them from interacting with repositories, such as opening or commenting on pull requests or issues. Learn more about blocking a user.