Keep no secret-shaped question, and hold the two-question cap across nodes (issue 589, review 105)

A question was kept on the bus before the router judged its words, so a token in
it reached a bucket every node reads: every word, the options' too, is now held
to the secret shapes and invisible characters before anything is kept, and a
refused question keeps none of its words. The cap was check-then-create and
failed open: each open question now holds one of two slot keys taken by create
or compare-and-set, refusing when a slot cannot be read. A question the router
never recorded frees its slot after two minutes instead of at expiry. by is
required; wording matches ADR 0332.

modules/claude-code/README.md

@@ -68,7 +68,7 @@ must see, a node that joins later included — kept, so it carries no secret eit
 | 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 |
 | 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 |
+| 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 |
 | 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) |
 
 ## The mesh's tools first (hq ADR 0245)
@@ -164,17 +164,22 @@ router's record. The text is never logged.
 **A session asks the operator a question on Telegram** (hq ADR 0332, issue 589). An agent that needs a decision while
 the operator may be away from the terminal calls `claude_code_ask` with a headline, the context in plain words, two to
 four options (a label and what choosing it means, exactly one recommended, with why), what it does if nobody
-answers, its name as who@node, and an expiry of 5 to 1430 minutes (60 when absent). The ask is published under this
+answers, its name as who@node (required), and an expiry of 5 to 1430 minutes (60 when absent). The ask is published under this
 module's name on `operator-channel`; every option is at the level approve and binds the digest of the question and of
-itself, so only a channel that proves who answered (Telegram) carries it, never the desk. The tool waits a few seconds
-for the router's word and answers with the question's key, or the router's refusal (the content rule: no path,
+itself, so only a channel that proves who answered (Telegram) carries it, never the desk. Anything shaped like a
+secret, or a character a phone does not show, in any of its words is refused before anything is kept. The tool waits
+a few seconds for the router's word and answers within seconds with the question's key, or the router's refusal (the content rule: no path,
 address, host name, commit id or secret's shape; `messenger_check` tests words first). `claude_code_answer` with the
 key says waiting, chosen (the option, who answered, on which channel, when), ended without a choice (expired,
 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; a session reads a question at most once a
-minute. At most two questions wait at once, mesh-wide, so they never take the module's last open ask from a
-proposal. An answer is the operator's word to the session and authorises nothing; a question never defaults.
+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 two minutes
+after it was asked (its refusal not heard here) frees its slot and reads as ended. 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.
 
 The home scope is registered through the tools **only where the home belongs to an agent account of its own**
 (hq ADR 0266, issue 340), and only for the node whose module serves the call. Where the agent runs as the operator,

modules/claude-code/cmd/claude-code/proposal_tools.go

@@ -22,16 +22,16 @@ type Proposing struct {
 	Now    func() time.Time
 	Write  WriteManaged
 	// Questions is the `questions` state (novox/hq ADR 0332); nil is the bus's, made on first use.
-	Questions   ProposalStore
+	Questions   QuestionStore
 	questionsMu sync.Mutex
 }
 
 // questions is the `questions` state.
-func (d *Proposing) questions() ProposalStore {
+func (d *Proposing) questions() QuestionStore {
 	d.questionsMu.Lock()
 	defer d.questionsMu.Unlock()
 	if d.Questions == nil {
-		d.Questions = stateOf{stdio.State(QuestionsState)}
+		d.Questions = keptQuestions{stdio.State(QuestionsState)}
 	}
 	return d.Questions
 }
@@ -154,11 +154,13 @@ func questionTools(p Paths, notLaunched error) []stdio.Tool {
 		{Name: "claude_code_ask",
 			Description: "Ask the operator a question on Telegram when you need a decision and the operator may not be at the " +
 				"terminal (novox/hq ADR 0332): a headline, the context in plain words, two to four options with exactly one " +
-				"marked recommended and why, what you do if nobody answers, and when it expires (5 to 1430 minutes, 60 when " +
-				"absent). Answers at once with the question's key; read the answer with claude_code_answer and keep working " +
+				"marked recommended and why, what you do if nobody answers, who you are (by, required), and when it expires (5 to " +
+				"1430 minutes, 60 when absent). Answers within seconds, once the router has taken it, with the question's " +
+				"key, or with the router's refusal; read the answer with claude_code_answer and keep working " +
 				"meanwhile. Only a channel that proves who answered carries it, never the desk. The words leave the mesh, so the " +
 				"router refuses a path, an address, a host name, a commit id or anything shaped like a secret: say it in " +
-				"words (messenger_check tests words first). At most two questions wait at once, mesh-wide. The answer is the " +
+				"words (messenger_check tests words first); anything shaped like a secret is refused before it is kept. At most two " +
+				"questions wait at once, mesh-wide. The answer is the " +
 				"operator's word to you; it authorises no call that needs the operator's approval.",
 			Input: map[string]any{
 				"headline": str("what is decided, in a few plain words, at most 60 characters, e.g. 'Merge the resolver fix tonight?'"),
@@ -171,7 +173,7 @@ func questionTools(p Paths, notLaunched error) []stdio.Tool {
 					}, "required": []string{"label", "means"}}},
 				"why":       str("why the recommended option is the one you recommend"),
 				"on_expiry": str("what you do if nobody answers before it expires, e.g. 'I leave the pull request open'"),
-				"by":        str("your session's name, as who@node or who, e.g. fix-589@g14"),
+				"by":        str("required: your session's name, as who@node or who, e.g. fix-589@g14"),
 				"expires_minutes": map[string]any{"type": "integer",
 					"description": "minutes until it expires, 5 to 1430; 60 when absent"},
 			},
@@ -191,7 +193,7 @@ func questionTools(p Paths, notLaunched error) []stdio.Tool {
 				"with the option, who answered, on which channel and when; ended without a choice (expired, refused, " +
 				"cancelled, replaced) with the router's words; or not proven, when the router's record does not hold to the " +
 				"question asked. Read from the router's own record. wait_seconds (at most 60) waits for an answer within the " +
-				"call: read at most once a minute.",
+				"call; read a question at most once a minute (asked of you, not enforced).",
 			Input: map[string]any{
 				"key":          str("the key claude_code_ask gave, e.g. q-0123456789abcdef"),
 				"wait_seconds": map[string]any{"type": "integer", "description": "wait up to this many seconds (at most 60) while it is waiting"},

modules/claude-code/cmd/claude-code/questions.go

@@ -27,6 +27,7 @@ import (
 	"sync"
 	"time"
 
+	stdio "git.novox.be/novox/mesh-sdk/go"
 	"git.novox.be/novox/mesh-sdk/go/asks"
 )
 
@@ -37,9 +38,25 @@ const QuestionsState = "questions"
 const QuestionPrefix = "q-"
 
 // MostQuestions is how many questions may wait at once, mesh-wide: the router allows an asker three open asks
-// (asks.MostOpen), and one is kept for a proposed instruction section (ADR 0274), so a proposal always finds room.
+// (asks.MostOpen), and questions alone never fill them, so one is left for a proposed instruction section (ADR 0274)
+// while two questions are open. Each open question holds one slot key in the `questions` state (slotKey), taken by
+// create or by compare-and-set, so two nodes asking at once cannot both take the last one.
 const MostQuestions = asks.MostOpen - 1
 
+// RouterHoldsWithin is how long after a question is asked the router has recorded it: it records an ask before it
+// takes its work. A question it holds no record of after that was refused (and the refusal not heard here) or never
+// taken: it no longer holds a slot, and reads as ended unless a record of it appears after all.
+var RouterHoldsWithin = 2 * time.Minute
+
+// slotKey is the key of one of the MostQuestions slots: not a question's key, so no question is ever read from it.
+func slotKey(n int) string { return fmt.Sprintf("slot-%d", n) }
+
+// slot is what a slot key holds: the question holding it.
+type slot struct {
+	Question string    `json:"question"`
+	At       time.Time `json:"at"`
+}
+
 // The bounds of a question's life (ADR 0332 §2): at least five minutes, at most a day less ten minutes (the
 // router's bound for an approval, with the margin a proposal keeps), an hour when none is given.
 const (
@@ -71,6 +88,35 @@ type QuestionOption struct {
 	Recommended bool   `json:"recommended,omitempty"`
 }
 
+// QuestionStore is the `questions` state: a value with its revision, created only when absent, updated only at the
+// revision read.
+type QuestionStore interface {
+	Get(key string) (json.RawMessage, uint64, bool, error)
+	Create(key string, value any) error
+	Update(key string, value any, revision uint64) error
+	Put(key string, value any) error
+}
+
+// keptQuestions is the `questions` state on the bus.
+type keptQuestions struct{ s stdio.KeptState }
+
+func (k keptQuestions) Get(key string) (json.RawMessage, uint64, bool, error) {
+	e, err := k.s.Get(key)
+	if err != nil || e == nil {
+		return nil, 0, false, err
+	}
+	return e.Value, e.Revision, true, nil
+}
+func (k keptQuestions) Create(key string, value any) error {
+	_, err := k.s.Create(key, value)
+	return err
+}
+func (k keptQuestions) Update(key string, value any, revision uint64) error {
+	_, err := k.s.Update(key, value, revision)
+	return err
+}
+func (k keptQuestions) Put(key string, value any) error { _, err := k.s.Put(key, value); return err }
+
 // Question is one question, as the `questions` state keeps it. Every field the ask is composed from is here, so the
 // answer is held to exactly the question asked.
 type Question struct {
@@ -205,11 +251,8 @@ func QuestionProblem(q Question) string {
 	if life := q.Expires.Sub(q.At); life < QuestionShortest || life > QuestionLongest {
 		add("it lasts %s; a question lasts %s to %s", life.Round(time.Minute), QuestionShortest, QuestionLongest)
 	}
-	for _, text := range []string{q.Headline, q.Context, q.Why, q.OnExpiry} {
-		if what := invisible(text); what != "" {
-			add("its words hold %s, which a phone does not show", what)
-			break
-		}
+	if problem := questionWordsProblem(q); problem != "" {
+		add("%s", problem)
 	}
 	if len(problems) == 0 {
 		return ""
@@ -217,6 +260,30 @@ func QuestionProblem(q Question) string {
 	return strings.Join(problems, "; ")
 }
 
+// questionWordsProblem says why a question's words may not be kept at all, or "": anything shaped like a secret, or a
+// character a phone does not show, in any word the operator is shown — the headline, the context, why, what happens
+// on expiry, and each option's label and meaning. Judged before anything is kept: the `questions` state is read by
+// every node, and the router judges the content rule only after the question was kept.
+func questionWordsProblem(q Question) string {
+	fields := []struct{ name, text string }{{"the headline", q.Headline}, {"the context", q.Context}, {"why", q.Why},
+		{"what happens on expiry", q.OnExpiry}}
+	for i, o := range q.Options {
+		fields = append(fields, struct{ name, text string }{fmt.Sprintf("option %d's label", i+1), o.Label},
+			struct{ name, text string }{fmt.Sprintf("what option %d means", i+1), o.Means})
+	}
+	for _, f := range fields {
+		for _, re := range secretShaped {
+			if re.MatchString(f.text) {
+				return f.name + " holds something shaped like a secret, which is never kept or shown on a phone"
+			}
+		}
+		if what := invisible(f.text); what != "" {
+			return f.name + " holds " + what + ", which a phone does not show"
+		}
+	}
+	return ""
+}
+
 func newQuestionID() string {
 	var b [8]byte
 	_, _ = rand.Read(b[:])
@@ -234,9 +301,9 @@ type QuestionInput struct {
 	Expires  time.Duration
 }
 
-// AskQuestion keeps a question and asks the operator. It answers at once with the question's key; the answer is read
-// with ReadAnswer.
-func AskQuestion(p Paths, in QuestionInput, store ProposalStore, bus AskBus, now time.Time) (map[string]any, error) {
+// AskQuestion keeps a question and asks the operator. It answers within seconds, once the router has said whether it
+// took the question, with the question's key; the answer is read with ReadAnswer.
+func AskQuestion(p Paths, in QuestionInput, store QuestionStore, bus AskBus, now time.Time) (map[string]any, error) {
 	refuse := func(why string) (map[string]any, error) { return nil, errors.New(why + ". Nothing was asked") }
 	by := strings.TrimSpace(in.By)
 	if who, node, found := strings.Cut(by, "@"); found {
@@ -246,7 +313,7 @@ func AskQuestion(p Paths, in QuestionInput, store ProposalStore, bus AskBus, now
 		by = who
 	}
 	if by == "" {
-		by = "unnamed"
+		return refuse("it does not say who asks: give by, your session's name as who@node, such as fix-589@g14")
 	}
 	life := in.Expires
 	if life == 0 {
@@ -262,21 +329,25 @@ func AskQuestion(p Paths, in QuestionInput, store ProposalStore, bus AskBus, now
 	if problem := QuestionProblem(q); problem != "" {
 		return refuse(problem)
 	}
-	if open := openQuestions(store, bus, now); len(open) >= MostQuestions {
-		return refuse(fmt.Sprintf("%d questions already wait for the operator's answer (%s), and sessions hold at most %d at "+
-			"once so a proposed instruction section always finds room; read them with claude_code_answer, or wait for "+
-			"one to end", len(open), strings.Join(open, ", "), MostQuestions))
-	}
 	ask := q.Ask()
 	if err := ask.Check(now); err != nil {
 		return refuse(err.Error())
 	}
+	// The question is kept before its slot is taken, so a slot always names a question another node can judge.
 	if err := store.Create(q.ID, q); err != nil {
 		return nil, fmt.Errorf("the question could not be kept, so the operator was not asked: %w", err)
 	}
+	if open, err := takeSlot(store, bus, q, now); err != nil {
+		_ = store.Put(q.ID, tombstone(q, "not asked: "+err.Error()))
+		return refuse(err.Error())
+	} else if open != nil {
+		_ = store.Put(q.ID, tombstone(q, "not asked: two questions were open"))
+		return refuse(fmt.Sprintf("%d questions already wait for the operator's answer (%s), and sessions hold at most %d "+
+			"at once, so one of the router's three open asks is left for a proposed instruction section; read them with "+
+			"claude_code_answer, or wait for one to end", len(open), strings.Join(open, ", "), MostQuestions))
+	}
 	if err := bus.Publish(asks.AskSubject(Module), ask, ask.ID); err != nil {
-		q.Asked = "not asked: " + err.Error()
-		_ = store.Put(q.ID, q)
+		_ = store.Put(q.ID, tombstone(q, "not asked: "+err.Error()))
 		return nil, fmt.Errorf("the operator could not be asked: %v", err)
 	}
 	say("question %s from the session %s asked of the operator", q.ID, q.By)
@@ -310,43 +381,99 @@ func labelsOf(q Question) string {
 	return strings.Join(out, ", ")
 }
 
-// openQuestions are the keys of the questions still waiting: asked, not ended, not expired.
-func openQuestions(store ProposalStore, bus AskBus, now time.Time) []string {
-	keys, err := store.Keys()
+// questionOpen says whether a question still waits: kept, asked, not ended, not expired, and held by the router or
+// asked less than RouterHoldsWithin ago. An unreadable record is open: a slot is never freed on a guess.
+func questionOpen(store QuestionStore, bus AskBus, id string, now time.Time) (bool, error) {
+	q, ok, err := readQuestionErr(store, id)
 	if err != nil {
-		return nil
+		return true, err
 	}
+	if !ok || q.Asked != "" || q.Ended != "" || !now.Before(q.Expires) {
+		return false, nil
+	}
+	entry, err := bus.Record(q.ID)
+	if err != nil {
+		return true, err
+	}
+	if entry == nil {
+		return now.Sub(q.At) < RouterHoldsWithin, nil
+	}
+	var rec routerRecord
+	if json.Unmarshal(entry.Value, &rec) != nil {
+		return true, nil
+	}
+	return rec.State == "" || rec.State == "open", nil
+}
+
+// takeSlot takes one of the MostQuestions slots for q: a slot no key holds, by create, or one whose question no longer
+// waits, by compare-and-set at the revision read. Of two nodes taking the same slot only one wins; the other tries
+// the next. It answers the open questions when every slot is held, and an error — refusing, never guessing — when a
+// slot or its question cannot be read.
+func takeSlot(store QuestionStore, bus AskBus, q Question, now time.Time) ([]string, error) {
+	mine := slot{Question: q.ID, At: q.At}
 	var open []string
-	for _, k := range keys {
-		q, ok := readQuestion(store, k)
-		if !ok || q.Asked != "" || q.Ended != "" || !now.Before(q.Expires) {
-			continue
+	for n := 1; n <= MostQuestions; n++ {
+		key := slotKey(n)
+		raw, rev, found, err := store.Get(key)
+		if err != nil {
+			return nil, fmt.Errorf("the open questions could not be counted (%v), so none more is asked", err)
 		}
-		if entry, err := bus.Record(q.ID); err == nil && entry != nil {
-			var rec routerRecord
-			if json.Unmarshal(entry.Value, &rec) == nil && rec.State != "" && rec.State != "open" {
+		if !found {
+			if err := store.Create(key, mine); err == nil {
+				return nil, nil
+			} else if !errors.Is(err, stdio.ErrChanged) {
+				return nil, fmt.Errorf("the open questions could not be counted (%v), so none more is asked", err)
+			}
+			if raw, rev, found, err = store.Get(key); err != nil || !found {
+				return nil, fmt.Errorf("the open questions could not be counted, so none more is asked")
+			}
+		}
+		var held slot
+		if json.Unmarshal(raw, &held) == nil && held.Question != "" {
+			waits, err := questionOpen(store, bus, held.Question, now)
+			if err != nil {
+				return nil, fmt.Errorf("the open questions could not be counted (%v), so none more is asked", err)
+			}
+			if waits {
+				open = append(open, held.Question)
 				continue
 			}
 		}
-		open = append(open, q.ID)
+		if err := store.Update(key, mine, rev); err == nil {
+			return nil, nil
+		} else if !errors.Is(err, stdio.ErrChanged) {
+			return nil, fmt.Errorf("the open questions could not be counted (%v), so none more is asked", err)
+		}
+		open = append(open, "one taken just now")
 	}
 	sort.Strings(open)
-	return open
+	return open, nil
 }
 
-func readQuestion(store ProposalStore, key string) (Question, bool) {
+// tombstone is what is kept of a question that was not asked or that the router refused: its key, who asked, when, and
+// why it ended — none of its words, which the router judged unfit to leave the mesh or never saw.
+func tombstone(q Question, ended string) Question {
+	return Question{ID: q.ID, By: q.By, From: q.From, At: q.At, Expires: q.Expires, Ended: ended}
+}
+
+func readQuestion(store QuestionStore, key string) (Question, bool) {
+	q, ok, _ := readQuestionErr(store, key)
+	return q, ok
+}
+
+func readQuestionErr(store QuestionStore, key string) (Question, bool, error) {
 	if !strings.HasPrefix(key, QuestionPrefix) {
-		return Question{}, false
+		return Question{}, false, nil
 	}
-	raw, found, err := store.Get(key)
+	raw, _, found, err := store.Get(key)
 	if err != nil || !found {
-		return Question{}, false
+		return Question{}, false, err
 	}
 	var q Question
 	if json.Unmarshal(raw, &q) != nil || q.ID != key {
-		return Question{}, false
+		return Question{}, false, nil
 	}
-	return q, true
+	return q, true, nil
 }
 
 // The router's word that a question ended without a choice, as heard on the decided subject.
@@ -369,9 +496,10 @@ func forgetQuestionEnd(id string) {
 }
 
 // HeardQuestionEnd takes the router's word that a question ended without a choice — refused, expired, cancelled or
-// replaced — and keeps it on the question, because a refused ask leaves no record of the router's. A choice is
-// never taken from this word: only from the router's record, held to the question (ReadAnswer).
-func HeardQuestionEnd(store ProposalStore, w asks.Warrant) {
+// replaced — and keeps it on the question, because a refused ask leaves no record of the router's. A refused question
+// keeps none of its words (tombstone). A choice is never taken from this word: only from the router's record, held to
+// the question (ReadAnswer).
+func HeardQuestionEnd(store QuestionStore, w asks.Warrant) {
 	if w.Asker != Module || w.Outcome == asks.OutcomeChosen || !strings.HasPrefix(w.Ask, QuestionPrefix) {
 		return
 	}
@@ -380,9 +508,14 @@ func HeardQuestionEnd(store ProposalStore, w asks.Warrant) {
 		return
 	}
 	if q.Ended == "" {
-		q.Ended = string(w.Outcome)
+		ended := string(w.Outcome)
 		if w.Words != "" {
-			q.Ended += ": " + w.Words
+			ended += ": " + w.Words
+		}
+		if w.Outcome == asks.OutcomeRefused {
+			q = tombstone(q, ended)
+		} else {
+			q.Ended = ended
 		}
 		_ = store.Put(q.ID, q)
 	}
@@ -401,7 +534,7 @@ const (
 )
 
 // ReadAnswer is where a question stands: waiting, chosen (and what), ended without a choice, or not proven.
-func ReadAnswer(store ProposalStore, bus AskBus, key string, now time.Time) (map[string]any, error) {
+func ReadAnswer(store QuestionStore, bus AskBus, key string, now time.Time) (map[string]any, error) {
 	q, ok := readQuestion(store, strings.TrimSpace(key))
 	if !ok {
 		return nil, fmt.Errorf("no question %q is kept: the key is the one claude_code_ask gave", key)
@@ -426,8 +559,9 @@ func ReadAnswer(store ProposalStore, bus AskBus, key string, now time.Time) (map
 	switch {
 	case entry == nil && q.Ended != "":
 		out["state"], out["words"] = AnswerEnded, q.Ended
-	case entry == nil && now.After(q.Expires.Add(time.Hour)):
-		out["state"], out["words"] = AnswerEnded, "the router keeps no record of it: refused before it was taken, or forgotten"
+	case entry == nil && now.Sub(q.At) >= RouterHoldsWithin:
+		out["state"], out["words"] = AnswerEnded, "the router holds no record of it: refused (its refusal was not heard "+
+			"here) or never taken; nothing waits for the operator's answer"
 	case entry == nil, rec.State == "open", rec.Warrant == nil:
 		out["state"] = AnswerWaiting
 		if q.Ended != "" {
@@ -472,7 +606,7 @@ func ReadAnswer(store ProposalStore, bus AskBus, key string, now time.Time) (map
 }
 
 // WaitForAnswer reads the answer, and while it is waiting reads it again until wait has passed.
-func WaitForAnswer(store ProposalStore, bus AskBus, key string, wait time.Duration, now func() time.Time) (map[string]any, error) {
+func WaitForAnswer(store QuestionStore, bus AskBus, key string, wait time.Duration, now func() time.Time) (map[string]any, error) {
 	if wait > AnswerWaitsAtMost {
 		wait = AnswerWaitsAtMost
 	}

modules/claude-code/cmd/claude-code/questions_test.go

@@ -6,10 +6,13 @@ package main
 
 import (
 	"encoding/json"
+	"errors"
 	"strings"
+	"sync"
 	"testing"
 	"time"
 
+	stdio "git.novox.be/novox/mesh-sdk/go"
 	"git.novox.be/novox/mesh-sdk/go/asks"
 )
 
@@ -27,7 +30,7 @@ func mergeTonight() QuestionInput {
 	}
 }
 
-func askMergeTonight(t *testing.T, p Paths, store *memProposals, r *router, at time.Time) string {
+func askMergeTonight(t *testing.T, p Paths, store *memQuestions, r *router, at time.Time) string {
 	t.Helper()
 	out, err := AskQuestion(p, mergeTonight(), store, r, at)
 	if err != nil {
@@ -38,7 +41,7 @@ func askMergeTonight(t *testing.T, p Paths, store *memProposals, r *router, at t
 
 func TestAQuestionIsAskedUnderTheModulesNameAtTheLevelApprove(t *testing.T) {
 	p, _ := node(t, "g14")
-	store, r := &memProposals{m: map[string]json.RawMessage{}}, newRouter()
+	store, r := newMemQuestions(), newRouter()
 	out, err := AskQuestion(p, mergeTonight(), store, r, t0)
 	if err != nil {
 		t.Fatal(err)
@@ -78,7 +81,7 @@ func TestAQuestionIsAskedUnderTheModulesNameAtTheLevelApprove(t *testing.T) {
 
 func TestTheChosenAnswerIsReadBackByItsKey(t *testing.T) {
 	p, _ := node(t, "g14")
-	store, r := &memProposals{m: map[string]json.RawMessage{}}, newRouter()
+	store, r := newMemQuestions(), newRouter()
 	key := askMergeTonight(t, p, store, r, t0)
 	out, err := ReadAnswer(store, r, key, t0.Add(time.Minute))
 	if err != nil || out["state"] != AnswerWaiting {
@@ -100,7 +103,7 @@ func TestTheChosenAnswerIsReadBackByItsKey(t *testing.T) {
 
 func TestAnAnswerThatDoesNotHoldToTheQuestionIsNotProven(t *testing.T) {
 	p, _ := node(t, "g14")
-	store, r := &memProposals{m: map[string]json.RawMessage{}}, newRouter()
+	store, r := newMemQuestions(), newRouter()
 	key := askMergeTonight(t, p, store, r, t0)
 	w := r.chose(key, "o1", t0.Add(time.Minute))
 
@@ -115,7 +118,7 @@ func TestAnAnswerThatDoesNotHoldToTheQuestionIsNotProven(t *testing.T) {
 	}
 
 	// A warrant for another ask's digest.
-	store2, r2 := &memProposals{m: map[string]json.RawMessage{}}, newRouter()
+	store2, r2 := newMemQuestions(), newRouter()
 	key2 := askMergeTonight(t, p, store2, r2, t0)
 	w2 := r2.chose(key2, "o1", t0.Add(time.Minute))
 	w2.AskDigest = "sha256:0000"
@@ -125,7 +128,7 @@ func TestAnAnswerThatDoesNotHoldToTheQuestionIsNotProven(t *testing.T) {
 	}
 
 	// A choice made after the question expired.
-	store3, r3 := &memProposals{m: map[string]json.RawMessage{}}, newRouter()
+	store3, r3 := newMemQuestions(), newRouter()
 	key3 := askMergeTonight(t, p, store3, r3, t0)
 	r3.close(r3.chose(key3, "o1", t0.Add(QuestionDefault+time.Minute)))
 	if out, _ := ReadAnswer(store3, r3, key3, t0.Add(2*QuestionDefault)); out["state"] != AnswerNotProven {
@@ -150,8 +153,19 @@ func TestAQuestionIsRefusedOutsideItsBounds(t *testing.T) {
 		"too long":          func(in *QuestionInput) { in.Expires = 2 * 24 * time.Hour },
 		"another node":      func(in *QuestionInput) { in.By = "fix-589@shanks" },
 		"an invisible char": func(in *QuestionInput) { in.Context += "\u200b" },
+		"invisible label":   func(in *QuestionInput) { in.Options[0].Label = "Merge\u200bnow" },
+		"invisible means":   func(in *QuestionInput) { in.Options[1].Means += "\u202e" },
+		"a token, context":  func(in *QuestionInput) { in.Context += " The key is ghp_abcdefghijklmnopqrstuvwxyz0123456789." },
+		"a token, headline": func(in *QuestionInput) { in.Headline = "Use sk-abcdefghijklmnopqrstuv?" },
+		"a token, why":      func(in *QuestionInput) { in.Why = "password: hunter2 works" },
+		"a token, expiry":   func(in *QuestionInput) { in.OnExpiry = "I keep AKIAABCDEFGHIJKLMNOP" },
+		"a token, label":    func(in *QuestionInput) { in.Options[0].Label = "token=abc" },
+		"a token, means":    func(in *QuestionInput) { in.Options[1].Means = "-----BEGIN PRIVATE KEY-----" },
+		"no by":             func(in *QuestionInput) { in.By = "" },
+		"61 characters":     func(in *QuestionInput) { in.Headline = strings.Repeat("x", 60) + "?" },
+		"1431 minutes":      func(in *QuestionInput) { in.Expires = 1431 * time.Minute },
 	} {
-		store, r := &memProposals{m: map[string]json.RawMessage{}}, newRouter()
+		store, r := newMemQuestions(), newRouter()
 		in := mergeTonight()
 		in.Options = append([]QuestionOption(nil), in.Options...)
 		change(&in)
@@ -166,7 +180,7 @@ func TestAQuestionIsRefusedOutsideItsBounds(t *testing.T) {
 
 func TestAtMostTwoQuestionsWaitSoAProposalFindsRoom(t *testing.T) {
 	p, _ := node(t, "g14")
-	store, r := &memProposals{m: map[string]json.RawMessage{}}, newRouter()
+	store, r := newMemQuestions(), newRouter()
 	first := askMergeTonight(t, p, store, r, t0)
 	askMergeTonight(t, p, store, r, t0)
 	if _, err := AskQuestion(p, mergeTonight(), store, r, t0); err == nil || !strings.Contains(err.Error(), "2 questions already wait") {
@@ -187,7 +201,7 @@ func TestAtMostTwoQuestionsWaitSoAProposalFindsRoom(t *testing.T) {
 // subject, with no record.
 type questionRefusingRouter struct {
 	*router
-	store ProposalStore
+	store QuestionStore
 	id    string
 }
 
@@ -203,7 +217,7 @@ func (r *questionRefusingRouter) Publish(subject string, body any, id string) er
 
 func TestTheRoutersRefusalIsSaidAndKept(t *testing.T) {
 	p, _ := node(t, "g14")
-	store := &memProposals{m: map[string]json.RawMessage{}}
+	store := newMemQuestions()
 	refusing := &questionRefusingRouter{router: newRouter(), store: store}
 	within, every := RouterAnswersWithin, RouterAnswersEvery
 	RouterAnswersWithin, RouterAnswersEvery = time.Second, 10*time.Millisecond
@@ -216,14 +230,19 @@ func TestTheRoutersRefusalIsSaidAndKept(t *testing.T) {
 	if err != nil || out["state"] != AnswerEnded || !strings.Contains(out["words"].(string), "refused") {
 		t.Errorf("read after the refusal: %v %v", out, err)
 	}
-	if open := openQuestions(store, refusing, t0.Add(time.Minute)); len(open) != 0 {
-		t.Errorf("a refused question still counts as waiting: %v", open)
+	if waits, err := questionOpen(store, refusing, refusing.id, t0.Add(time.Minute)); waits || err != nil {
+		t.Errorf("a refused question still counts as waiting: %v %v", waits, err)
+	}
+	var kept Question
+	_ = json.Unmarshal(store.m[refusing.id], &kept)
+	if kept.Headline != "" || kept.Context != "" || len(kept.Options) != 0 || kept.Why != "" || kept.OnExpiry != "" {
+		t.Errorf("a refused question keeps its words: %+v", kept)
 	}
 }
 
 func TestAnExpiredQuestionSaysWhatTheSessionDoes(t *testing.T) {
 	p, _ := node(t, "g14")
-	store, r := &memProposals{m: map[string]json.RawMessage{}}, newRouter()
+	store, r := newMemQuestions(), newRouter()
 	key := askMergeTonight(t, p, store, r, t0)
 	r.close(asks.Warrant{Ask: key, Outcome: asks.OutcomeExpired, At: t0.Add(QuestionDefault)})
 	out, err := ReadAnswer(store, r, key, t0.Add(QuestionDefault+time.Minute))
@@ -235,7 +254,7 @@ func TestAnExpiredQuestionSaysWhatTheSessionDoes(t *testing.T) {
 func TestTheQuestionToolsAreServed(t *testing.T) {
 	p, _ := node(t, "g14")
 	r := newRouter()
-	questions := &memProposals{m: map[string]json.RawMessage{}}
+	questions := newMemQuestions()
 	proposing = &Proposing{Store: &memProposals{m: map[string]json.RawMessage{}}, Bus: r, Now: func() time.Time { return t0 },
 		Write: writer(map[string]string{}), Questions: questions}
 	t.Cleanup(func() { proposing = nil })
@@ -271,7 +290,7 @@ func TestAReadWaitsNoLongerThanItsBound(t *testing.T) {
 		t.Fatalf("a read may wait %s; ADR 0332 says 60 s", AnswerWaitsAtMost)
 	}
 	p, _ := node(t, "g14")
-	store, r := &memProposals{m: map[string]json.RawMessage{}}, newRouter()
+	store, r := newMemQuestions(), newRouter()
 	key := askMergeTonight(t, p, store, r, t0)
 	most, every := AnswerWaitsAtMost, AnswerLooksEvery
 	AnswerWaitsAtMost, AnswerLooksEvery = 100*time.Millisecond, 10*time.Millisecond
@@ -285,3 +304,165 @@ func TestAReadWaitsNoLongerThanItsBound(t *testing.T) {
 		t.Errorf("a read asked to wait an hour took %s", took)
 	}
 }
+
+// memQuestions is the `questions` state: values with revisions, create-only and compare-and-set, safe for the
+// goroutines a test starts.
+type memQuestions struct {
+	mu  sync.Mutex
+	m   map[string]json.RawMessage
+	rev map[string]uint64
+	n   uint64
+	// failGet fails every read of a key with this prefix.
+	failGet string
+}
+
+func newMemQuestions() *memQuestions {
+	return &memQuestions{m: map[string]json.RawMessage{}, rev: map[string]uint64{}}
+}
+
+func (s *memQuestions) Get(key string) (json.RawMessage, uint64, bool, error) {
+	s.mu.Lock()
+	defer s.mu.Unlock()
+	if s.failGet != "" && strings.HasPrefix(key, s.failGet) {
+		return nil, 0, false, errors.New("the bus did not answer")
+	}
+	raw, ok := s.m[key]
+	return raw, s.rev[key], ok, nil
+}
+func (s *memQuestions) put(key string, value any) error {
+	raw, err := json.Marshal(value)
+	s.n++
+	s.m[key], s.rev[key] = raw, s.n
+	return err
+}
+func (s *memQuestions) Create(key string, value any) error {
+	s.mu.Lock()
+	defer s.mu.Unlock()
+	if _, ok := s.m[key]; ok {
+		return stdio.ErrChanged
+	}
+	return s.put(key, value)
+}
+func (s *memQuestions) Update(key string, value any, revision uint64) error {
+	s.mu.Lock()
+	defer s.mu.Unlock()
+	if s.rev[key] != revision {
+		return stdio.ErrChanged
+	}
+	return s.put(key, value)
+}
+func (s *memQuestions) Put(key string, value any) error {
+	s.mu.Lock()
+	defer s.mu.Unlock()
+	return s.put(key, value)
+}
+
+// The edges of ADR 0332 §2 that are allowed: a 60-character headline and 1430 minutes.
+func TestTheEdgesOfAQuestionAreAsked(t *testing.T) {
+	p, _ := node(t, "g14")
+	in := mergeTonight()
+	in.Headline = strings.Repeat("x", 59) + "?"
+	in.Expires = 1430 * time.Minute
+	if _, err := AskQuestion(p, in, newMemQuestions(), newRouter(), t0); err != nil {
+		t.Fatalf("a 60-character headline lasting 1430 minutes: %v", err)
+	}
+}
+
+// A warrant naming another asker or another ask, or an option the question did not offer, is no answer to it.
+func TestAWarrantForAnotherAskOrOptionIsNotProven(t *testing.T) {
+	p, _ := node(t, "g14")
+	for name, change := range map[string]func(*asks.Warrant){
+		"another asker":       func(w *asks.Warrant) { w.Asker = "mesh-controller" },
+		"another ask":         func(w *asks.Warrant) { w.Ask = "q-0000000000000000" },
+		"an option unoffered": func(w *asks.Warrant) { w.Option, w.Label = "o9", "Delete everything" },
+	} {
+		store, r := newMemQuestions(), newRouter()
+		key := askMergeTonight(t, p, store, r, t0)
+		w := r.chose(key, "o1", t0.Add(time.Minute))
+		change(&w)
+		r.records[key], _ = json.Marshal(map[string]any{"state": "chosen", "warrant": w})
+		if out, _ := ReadAnswer(store, r, key, t0.Add(2*time.Minute)); out["state"] != AnswerNotProven {
+			t.Errorf("%s read as %v", name, out)
+		}
+	}
+}
+
+// Two nodes asking at the same moment take two slots at most: of three sessions racing, one is refused.
+func TestTheCapHoldsAcrossNodesAskingAtOnce(t *testing.T) {
+	store, r := newMemQuestions(), &lockedRouter{router: newRouter()}
+	nodes := []string{"g14", "shanks", "ace"}
+	var wg sync.WaitGroup
+	errs := make([]error, len(nodes))
+	for i, name := range nodes {
+		p, _ := node(t, name)
+		wg.Add(1)
+		go func(i int, p Paths) {
+			defer wg.Done()
+			in := mergeTonight()
+			in.By = "fix-589"
+			_, errs[i] = AskQuestion(p, in, store, r, t0)
+		}(i, p)
+	}
+	wg.Wait()
+	refused := 0
+	for _, err := range errs {
+		if err != nil {
+			if !strings.Contains(err.Error(), "questions already wait") {
+				t.Errorf("refused for another reason: %v", err)
+			}
+			refused++
+		}
+	}
+	if refused != 1 {
+		t.Errorf("%d of three questions asked at once were refused; one is", refused)
+	}
+}
+
+// A slot that cannot be read refuses the question: the cap fails closed.
+func TestTheCapFailsClosed(t *testing.T) {
+	p, _ := node(t, "g14")
+	store, r := newMemQuestions(), newRouter()
+	store.failGet = "slot-"
+	if _, err := AskQuestion(p, mergeTonight(), store, r, t0); err == nil || !strings.Contains(err.Error(), "could not be counted") {
+		t.Fatalf("asked with the slots unreadable: %v", err)
+	}
+	if len(r.subjects) != 0 {
+		t.Error("the operator was asked anyway")
+	}
+}
+
+// A question whose refusal was not heard here, and of which the router holds no record, frees its slot once the
+// router has had time to record it, and reads as ended — not open until it expires.
+func TestAQuestionTheRouterNeverHeldFreesItsSlot(t *testing.T) {
+	p, _ := node(t, "g14")
+	store, r := newMemQuestions(), newRouter()
+	first := askMergeTonight(t, p, store, r, t0)
+	askMergeTonight(t, p, store, r, t0)
+	delete(r.records, first) // refused, and the refusal never heard here
+	if _, err := AskQuestion(p, mergeTonight(), store, r, t0.Add(time.Minute)); err == nil {
+		t.Fatal("a slot was freed before the router had time to record the question")
+	}
+	if out, _ := ReadAnswer(store, r, first, t0.Add(RouterHoldsWithin)); out["state"] != AnswerEnded {
+		t.Errorf("read as %v", out)
+	}
+	if _, err := AskQuestion(p, mergeTonight(), store, r, t0.Add(RouterHoldsWithin)); err != nil {
+		t.Errorf("the slot of a question the router never held was not freed: %v", err)
+	}
+}
+
+// lockedRouter is the router fake, safe for sessions asking at once.
+type lockedRouter struct {
+	mu sync.Mutex
+	*router
+}
+
+func (l *lockedRouter) Publish(subject string, body any, id string) error {
+	l.mu.Lock()
+	defer l.mu.Unlock()
+	return l.router.Publish(subject, body, id)
+}
+func (l *lockedRouter) Record(id string) (*stdio.StateEntry, error) {
+	l.mu.Lock()
+	defer l.mu.Unlock()
+	return l.router.Record(id)
+}