mesh/merge-gate head 0f159a166200 merged into trunk 7a7a4034852a — pass: builds claude-code → ace, g14, novox, shanks; no bus step; every machine composes …
mesh/repo-check head 0f159a166200 merged into trunk 7a7a4034852a — pass: its merge-check.sh passed, each part in its toolchain: merge-check.sh (go), merge-…
The Stop hook treated a missing waits file exactly like one that could
not be read: a notice on every stop of every session with no waits —
the normal state, not an exceptional one. DecideStop now treats
os.ErrNotExist as the ordinary case of no open waits, silently; the
notice is still said for a file that exists but cannot be read or does
not parse.
ADR 0490 §1 keeps waits under a name, but the file a hook reads is kept
per session id: writing it from only the name an op just acted on meant
a second name bound to the same session id had its waits silently drop
out of enforcement the moment the first name was touched, and adopting
one name away emptied the id's file outright even when another name was
still bound to it. Every write now rewrites a session's file from the
union of every name currently bound to it (sessionWaitsUnion), adopt's
rewrite of the superseded id included.
Also: each atomic write gets its own temp file name, so two writes to
the same path can never interleave on one; adopt clears the refusal
counter of both the superseded and the adopting session, so a fresh
three-strikes count starts under the new binding; and list is read-only
again — it no longer claims an unbound name or refuses a session that
is not (yet) the one bound to it, and instead says who currently holds
the name.
Claude Code gives a session a new session_id at /clear, with nothing
naming the earlier one, so keying claude_code_wait's state by session id
lost a session's waits the moment it cleared. ADR 0490 changes ADR 0489
§1: waits are now kept under the caller's name and node (by, who@node);
every op also takes session, the caller's current id, and the module
keeps a binding of the name to the session that last called for it. An
unbound name is claimed by the first call for it; a session whose own
id no longer matches the binding is refused on every op, naming the
session that holds it, until it calls the new op adopt, which always
takes the name over at once (no bound protects a recent call) and tells
the earlier session so on its own next call.
The mirrored file a hook reads stays keyed by session id (a hook only
ever has that), holding whichever name's waits are currently bound to
it; adopting rewrites both the old id's file (emptied) and the new id's
(filled) in the same call. A second mirrored file, the node-wide index,
lists every name on the node with an open wait; the SessionStart hook
now also reads it, on every start, and prints the exact adopt call for
each name, so a session with a brand new id can find and reclaim its
own waits without calling back into the module.
A session id reaching the Stop and SessionStart hooks was only checked
for blank, so "../../x" or "a/b" could read or write outside the waits
directory; both hooks now hold every session id to waitSessionID before
building any path, and allow (saying why) rather than fail on a bad one.
claude_code_wait's list/watched/done/said held only add to that same
check, letting a malformed id through as an ordinary, always-empty key;
they all refuse it now.
The Stop hook's three-strikes cap could be bypassed forever if its
counter file could not be written on a refusal: it now allows instead,
saying so, so a wait can never hold a turn open indefinitely. Its advice
offered mesh_wait on a wait's event family unconditionally, ignoring
that the session's tool runner may not be granted it (ADR 0367 §2); it
now offers mesh_wait only "if this session's tool runner is granted it,
otherwise start a watcher", and drops the family entirely when a wait
has none.
Both hooks' fail-open and let-through notices were said only on stderr,
which Claude Code does not show the operator on an allowed (exit 0)
outcome; they now also print the same words as a "systemMessage" JSON
object on stdout, which Claude Code does show.
A session could end its turn while a wait it wrote down with
claude_code_wait had nothing following it. The plugin's Stop hook
(claude-code stop <waits-dir>) refuses the stop while a wait is within
its bound and unwatched, or past its bound and not said, reading the
waits file directly with no network and a 5s bound; it fails open on a
missing or malformed file or a panic, and lets a wait through after
three refusals in a row rather than holding the session forever.
Renames TestThePluginsOnlyHookIsTheGuard to cover the session-start and
Stop hooks too.
A session is never told its session id by any tool, only by a hook's
stdin, so it cannot name itself to claude_code_wait. The plugin's
SessionStart hook (claude-code session-start <waits-dir>) prints the id
and how to give it, plus the session's still-open waits after a /clear,
reading the waits file claude_code_wait mirrors directly rather than
calling back into this module. Carries the Stop hook's command and the
GuardFiles.WaitsDir plumbing MarketplaceWith needs for both (issue 1743
wires the Stop hook's dispatch next).
A session can end its turn while an Approve it expects is still open and
nothing then watches it. claude_code_wait (add/watched/done/said/list)
keeps each session's waits in the module's waits state, keyed by its
session id and node, and mirrors them whole and atomically to a file
under the agent's home that the plugin's coming hooks (1743, 1744) will
read without calling back into this module.
jochen