Compare commits

..

4 Commits

Author SHA1 Message Date
Dai Ha fe46311266 CB-604: reject an unrecognized profile kind at config load
CI / contract (pull_request) Successful in 53s
CI / build (pull_request) Successful in 1m10s
2026-08-16 18:37:21 +02:00
Dai Ha 27bbd11f06 Merge CB-584: carry the agent session id on a failed ticket's detail
CI / contract (push) Successful in 47s
CI / build (push) Successful in 1m23s
Most of issue #65 was already shipped in 5d5b3bd - MemberSession
records agentSessionId, PeerHandle.agentSessionId() has no default, the
roster exposes it, and bridge_spawn accepts sessionName and
resumeSessionId. That commit left one item for follow-up: carrying the
id on ReleaseDetail.

This does that item. CB-578 stage C already lets a lead re-dispatch onto
the same worktree after a failure; without the session id that is a cold
start. With it, the work and the thread both survive.

Also updates the bridge_spawn and bridge_list rows in the CLAUDE.md
intent table, which the earlier commit missed.

Verified here: trial merge onto main builds 830 tests BUILD SUCCESS,
unpiped. Confirmed against the code that items 1-4 really were already
on main, so the scope-down is correct rather than work skipped.

Closes #65
2026-08-16 18:29:24 +02:00
Dai Ha 48d7841fbf CB-589: document that weighted placement is not cheapest-first
CI / contract (push) Successful in 50s
CI / build (push) Successful in 1m43s
The weight ratio does not express a preference order. weighted spreads
spawns across every profile with a free slot, so paid spawns happen
while the free box is idle - and a profile at maxLoad freezes its score,
so it can lose the next pick after a slot frees.

The real fix is a cost-first policy (CB-589). This documents the
workaround and its trap next to the key, because bridged.yaml is
gitignored: a fresh host starts without the workaround and quietly pays,
with nothing to tell the operator why.

Comments only. BridgedConfigTest: 85 tests, BUILD SUCCESS.
2026-08-16 18:27:19 +02:00
Dai Ha cc1df11f69 CB-584: carry agentSessionId on a failed ticket's ReleaseDetail
CI / contract (pull_request) Successful in 1m6s
CI / build (pull_request) Successful in 1m42s
Closes the one piece issue #65 deliberately left out of 5d5b3bd: a
released member's agentSessionId now rides alongside worktree, branch
and snapshotRef on ReleaseDetail, and Bridged's onRelease handler
names it in the abandon reason, so a lead can resume the member's
conversation instead of only re-dispatching a fresh one onto the same
files.

Also updates the CLAUDE.md bridge_spawn/bridge_list table row, which
5d5b3bd shipped the sessionName/resumeSessionId/agentSessionId surface
for but never updated.
2026-08-16 18:25:24 +02:00
15 changed files with 193 additions and 606 deletions
+2 -2
View File
@@ -105,8 +105,8 @@ the merge — and merging on a reviewer's word is delegating it by proxy.
|---|---|
| Confirm your own role | `bridge_whoami` |
| See backends available | `bridge_profiles` |
| Start a member | `bridge_spawn{role?, profile?, cwd?, worktree?, ticket?}` → `sessionId` + `paneId` |
| See the fleet | `bridge_list` → `leads` (your peers) + `members` · one peer's state: `bridge_status{sessionId}` |
| Start a member | `bridge_spawn{role?, profile?, cwd?, worktree?, ticket?, sessionName?, resumeSessionId?}` → `sessionId` + `paneId` |
| See the fleet | `bridge_list` → `leads` (your peers) + `members` (each carries `agentSessionId` when its backend knows one) · one peer's state: `bridge_status{sessionId}` |
| Delegate (blocking) | `bridge_send{sessionId, content}` |
| Delegate (long task) | `bridge_send{sessionId, content, wait:false}` → ticket → `bridge_poll{ticket}` |
| Answer a member's `bridge_ask` | `bridge_send{turnId, content}` — **not** `sessionId` |
+20
View File
@@ -275,6 +275,26 @@ profiles:
# argv: ["opencode"]
# How an unqualified spawn chooses a profile: fixed (default, reproduces pre-CB-518 behaviour),
# round-robin, or weighted. Omitting this key is a strict no-op for existing configs.
#
# `weighted` IS NOT "cheapest first" — read this before you set weights (CB-589).
# It is smooth weighted round-robin: it spreads spawns across EVERY profile that has a free slot,
# in weight ratio. It has no idea which profile costs money. So with local:10 / paid:2 you do not
# get "use local, overflow to paid" — you get roughly one spawn in six going to the paid profile
# while the local box still has a free slot.
#
# There is a sharper second effect. The policy's running score map lives for the daemon's whole
# life. While a profile is at maxLoad it is filtered out and its score FREEZES, so the paid
# profiles keep accumulating against it. When the local slot frees up it returns with a stale
# score and can LOSE the next pick — a paid spawn while the free box sits idle.
#
# Until a real cost-first policy exists, the workaround is to make the ratio decisive rather than
# proportional: give the free profile a weight so large that it wins every pick it is eligible
# for, and paid profiles only ever take genuine overflow. On this host that is local weight 100
# against paid weights of ~1.
#
# The gotcha with that workaround: it expresses a PREFERENCE ORDER through a RATIO knob. Add a
# future profile at weight 150 and it silently outranks the free box, with nothing to warn you.
# Re-check the weights whenever you add a profile.
placement: weighted
# How long a credential sits out after a BACKEND_EXHAUSTED classification (CB-578 stage B), in
@@ -441,6 +441,11 @@ public final class Bridged {
reason += "; worktree=" + detail.worktreePath() + " branch=" + detail.branch()
+ " snapshot=" + (detail.snapshotRef() != null ? detail.snapshotRef() : "none");
}
// CB-584 (issue #65 criterion 5): also name the agent session, so a lead can resume the
// member's conversation instead of only re-dispatching a fresh one onto the same files.
if (detail.agentSessionId() != null) {
reason += " agentSessionId=" + detail.agentSessionId();
}
messages.abandon(detail.terminalId(), reason);
replyInbox.release(detail.terminalId());
primaryRegistry.forgetDelegation(detail.terminalId()); // CB-532: don't leak the lead binding
@@ -986,6 +986,7 @@ public record BridgedConfig(
warnUnknownTopLevelKeys(yaml, path);
rejectDuplicateMemberSlots(yaml);
rejectNegativeMaxLoad(yaml);
rejectUnknownKind(yaml);
BridgedConfig cfg = YAML.readValue(yaml, BridgedConfig.class);
return cfg.withDefaults();
} catch (IOException e) {
@@ -1280,6 +1281,53 @@ public record BridgedConfig(
}
}
/** The peer kinds this build has an adapter for — {@link Profile#kind()}'s only valid values. */
private static final Set<String> KNOWN_KINDS = Set.of(Profile.KIND_CLAUDE_CODE, Profile.KIND_OPENCODE);
/**
* Reject a profile whose {@code kind:} is not one of {@link #KNOWN_KINDS} (CB-604), naming the
* profile, the value it set, and the accepted set.
*
* <p>{@link Profile}'s compact constructor only lower-cases {@code kind} and compares it against
* {@code KIND_OPENCODE} — anything else, including a typo like {@code opencod}, silently falls
* into the claude-code bucket ({@link dev.ltms.bridged.member.CompositePeerLauncher} routes by
* exact adapter claim, not by membership in a known set). With {@code argv:} also unset, the argv
* default special-cases only the exact string {@code "claude-code"}, so the launch command falls
* back to {@code List.of(kind)} — the daemon then tries to run a program literally named after the
* typo. {@code CompositePeerLauncher}'s constructor already treats a profile claimed by two
* adapters as fatal (CB-402); an unrecognized kind is the same class of adapter-routing mistake
* and gets the same treatment here, at config load, rather than surfacing later as a failed spawn.
*
* @param yaml the raw config text
* @throws IllegalStateException when any profile's {@code kind} is a non-blank value not in
* {@link #KNOWN_KINDS} (case-insensitive)
*/
static void rejectUnknownKind(String yaml) {
Map<?, ?> raw;
try {
raw = YAML.readValue(yaml, Map.class);
} catch (IOException | IllegalArgumentException e) {
return; // a malformed file is reported by the real parse, not here
}
if (raw == null || !(raw.get("profiles") instanceof Map<?, ?> profiles)) {
return;
}
List<String> bad = profiles.entrySet().stream()
.filter(e -> e.getValue() instanceof Map<?, ?> p
&& p.get("kind") instanceof String k && !k.isBlank()
&& !KNOWN_KINDS.contains(k.toLowerCase()))
.map(e -> String.valueOf(e.getKey()) + "=" + ((Map<?, ?>) e.getValue()).get("kind"))
.sorted()
.toList();
if (!bad.isEmpty()) {
throw new IllegalStateException("refusing to start: profile(s) [" + String.join(", ", bad)
+ "] set an unrecognized kind — accepted values are "
+ String.join(", ", KNOWN_KINDS.stream().sorted().toList())
+ " (case-insensitive); an unrecognized kind would otherwise fall back to the"
+ " claude-code adapter and try to launch a program named after the typo.");
}
}
static List<String> unknownTopLevelKeys(String yaml) {
Map<?, ?> raw;
try {
@@ -577,25 +577,13 @@ public final class BridgeMcp {
return text("acknowledged " + msgId);
}
/**
* {@code bridge_status}: the live lifecycle status of a worker session, plus — when the worker
* is paused mid-turn in an async {@code bridge_ask} (CB-582) — the open question and how to
* answer it, so a lead on its normal poll cadence does not need the ticket to notice.
*/
/** {@code bridge_status}: the live lifecycle status of a worker session. */
static McpSchema.CallToolResult status(MessageService messages, String sessionId) {
if (isBlank(sessionId)) {
return error("sessionId is required");
}
try {
String base = messages.status(sessionId).name().toLowerCase();
MessageService.PendingAsk ask = messages.pendingAsk(sessionId);
if (ask == null) {
return text(base);
}
return text(base + "\n\n[question — worker is waiting for your answer]\n" + ask.question()
+ "\n\nAnswer it by calling bridge_send again with turnId=\"" + ask.turnId()
+ "\" and content set to your answer; the worker resumes the same turn."
+ " (ticket " + ask.ticket() + ")");
return text(messages.status(sessionId).name().toLowerCase());
} catch (HerdrException e) {
return error("herdr error for session " + sessionId + ": " + e.getMessage());
}
@@ -164,30 +164,18 @@ public final class MessageService {
/** An in-flight or finished async delegation, keyed by its ticket. */
private static final class Task {
private final String ticket;
private final String target;
private final CompletableFuture<Reply> future = new CompletableFuture<>();
private final long createdNanos;
private volatile Reply question;
private volatile String turnId;
private Task(String ticket, String target, long createdNanos) {
this.ticket = ticket;
private Task(String target, long createdNanos) {
this.target = target;
this.createdNanos = createdNanos;
}
}
/**
* A worker session's currently-open {@code bridge_ask} question, surfaced so {@code bridge_status}
* can show it without the caller needing the ticket first (CB-582). Only covers async
* (fire-and-poll) delegations, which track the question on their {@link Task}; a blocking
* ({@code wait:true}) send already hands the question straight back to its own caller, so there is
* nothing hidden left for {@code bridge_status} to surface in that case.
*/
public record PendingAsk(String ticket, String question, String turnId) {
}
private final AgentControl agents;
private final Injector injector;
private final Rendezvous rendezvous;
@@ -212,11 +200,9 @@ public final class MessageService {
* Create with an explicit {@link ReplyInbox} and optional {@link ReplyPushLoop}.
*
* @param pushLoop nullable — when non-null, the push loop is notified on the no-waiter reply
* branch ({@link #reply}) so it can nudge the primary to drain the inbox,
* branch ({@link #reply}) so it can nudge the primary to drain the inbox, and
* (CB-588) whenever an async ticket started by {@link #sendAsync} reaches a
* terminal phase, whenever {@link #poll} hands a terminal ticket to its caller,
* and (CB-582) whenever an async ticket's worker pauses mid-turn in
* {@code bridge_ask} or that pause ends (answered or lapsed)
* terminal phase, and whenever {@link #poll} hands a terminal ticket to its caller
*/
public MessageService(AgentControl agents, Injector injector, Rendezvous rendezvous,
ReplyInbox inbox, ReplyPushLoop pushLoop) {
@@ -509,15 +495,6 @@ public final class MessageService {
rendezvous.closeAsk(ticket.turnId());
return new AskResult(AskOutcome.NO_WAITER, null); // no primary is blocked on this worker
}
// CB-582: the question just became visible via bridge_poll (Phase.ASKING) for an async
// (wait:false) delegation — nudge the lead's own pane the same way a terminal ticket does
// (CB-588), since the lead's normal poll cadence is minutes away and the reverse-rendezvous
// window (~55s, see BridgeMcp/BridgedApp) is far shorter. A blocking (wait:true) send has
// no Task and gets the question directly in its own reply, so task == null there — nothing
// to nudge.
if (task != null && pushLoop != null) {
pushLoop.onQuestionOpened(task.ticket, workerSession, ticket.turnId(), question);
}
}
try {
String answer = ticket.answer().get(timeoutMillis, TimeUnit.MILLISECONDS);
@@ -612,7 +589,7 @@ public final class MessageService {
*/
public String sendAsync(String target, String content, Runnable onAccepted) {
String ticket = "task-" + ticketSeq.incrementAndGet();
Task task = new Task(ticket, target, nowNanos.getAsLong());
Task task = new Task(target, nowNanos.getAsLong());
tasks.put(ticket, task);
if (pushLoop != null) {
// CB-588: task.future only ever completes on a terminal phase (DONE or a failure) — a
@@ -738,12 +715,6 @@ public final class MessageService {
/** Clear an answered or lapsed question, but only when it matches the ticket's current turn. */
private void clearAsyncQuestion(String turnId, boolean forgetTurn) {
// CB-582: tell the push loop first — like ticketCollected, a removal for a turnId it never
// nudged about (or already dropped) is a harmless no-op, so this is safe to call unconditionally
// rather than threading the guard below through it.
if (pushLoop != null) {
pushLoop.questionClosed(turnId);
}
Task task = asyncTasksByTurn.get(turnId);
if (task != null && turnId.equals(task.turnId)) {
task.question = null;
@@ -775,23 +746,6 @@ public final class MessageService {
return asyncTasksByTurn.values().stream().anyMatch(task -> target.equals(task.target));
}
/**
* The question {@code workerSession} is currently paused on via {@code bridge_ask}, if any
* (CB-582) — {@code bridge_status} uses this to show a pending question without the caller
* needing the ticket. {@code null} when the session has no open async question (including a
* session mid a <em>blocking</em> {@code bridge_ask}, which has no {@link Task} to look up — see
* {@link PendingAsk}).
*/
public PendingAsk pendingAsk(String workerSession) {
for (Task task : tasks.values()) {
Reply q = task.question;
if (q != null && workerSession.equals(task.target)) {
return new PendingAsk(task.ticket, q.text(), q.turnId());
}
}
return null;
}
/** Release the async executor. */
public void close() {
asyncExecutor.shutdown();
@@ -19,33 +19,30 @@ import java.util.stream.Collectors;
/**
* A status-gated push loop that nudges a lead's own herdr pane when it has uncollected work
* waiting: a worker reply queued with no live {@code bridge_send} to resolve it (CB-307), an
* waiting: a worker reply queued with no live {@code bridge_send} to resolve it (CB-307), or an
* async delegation ticket ({@code bridge_send(wait:false)}) that reached a terminal phase
* (CB-588), or an async ticket's worker pausing mid-turn in {@code bridge_ask} to await an answer
* (CB-582).
* (CB-588).
*
* <p><strong>CB-590: one schedule per lead.</strong> All three kinds of work are triggered
* through their own entry point — {@link #onReplyQueued(String)},
* {@link #onTicketTerminal(String, String, boolean)}, and
* {@link #onQuestionOpened(String, String, String, String)} — but each resolves the lead that
* should be nudged and coalesces onto a single per-lead reminder schedule, tracked in
* {@link #activeLeads}. Earlier this was two independent schedules (one keyed by worker target
* for replies, one keyed by lead for tickets) that could both decide to inject into the same pane
* in the same window — a race, not routine behaviour, but the expensive kind: it interrupts the
* lead's live turn twice. Collapsing to one schedule per lead makes that structurally impossible:
* at most one scheduled tick chain is ever live for a given lead (guarded by {@link #activeLeads}'
* <p><strong>CB-590: one schedule per lead.</strong> Both kinds of work are triggered through
* their own entry point — {@link #onReplyQueued(String)} and
* {@link #onTicketTerminal(String, String, boolean)} — but both resolve the lead that should be
* nudged and coalesce onto a single per-lead reminder schedule, tracked in {@link #activeLeads}.
* Earlier this was two independent schedules (one keyed by worker target for replies, one keyed
* by lead for tickets) that could both decide to inject into the same pane in the same window —
* a race, not routine behaviour, but the expensive kind: it interrupts the lead's live turn
* twice. Collapsing to one schedule per lead makes that structurally impossible: at most one
* scheduled tick chain is ever live for a given lead (guarded by {@link #activeLeads}'
* compare-and-set), so at most one {@code agents.send} to that lead's pane is ever in flight.
*
* <p>Each tick examines <em>everything</em> pending for that lead — reply targets whose inbox
* still holds an unacked message ({@link #pendingReplies}), tickets not yet collected
* ({@link #pendingTickets}), and open questions not yet answered or lapsed
* ({@link #pendingQuestions}) — and sends at most one combined nudge per tick
* ({@link #injectNudge(String, int, int, int)}). Work that arrives while the lead is busy is
* never lost: it is re-read fresh on every tick until the lead is injectable or its own reminder
* cap ({@link #maxReminders}) is reached — each source spends from its own budget, so one source
* exhausting its cap does not stop nudges about the others (post-CB-590 regression fix; see
* {@link #decide}) — whichever the durable inbox / pending set doesn't already answer via
* {@code STOP}.
* still holds an unacked message ({@link #pendingReplies}) and tickets not yet collected
* ({@link #pendingTickets}) — and sends at most one combined nudge per tick
* ({@link #injectNudge(String, int, int)}). Work that arrives while the lead is busy is never
* lost: it is re-read fresh on every tick until the lead is injectable or its own reminder cap
* ({@link #maxReminders}) is reached — reply and ticket work each spend from their own budget, so
* one source exhausting its cap does not stop nudges about the other (post-CB-590 regression fix;
* see {@link #decide}) — whichever the durable inbox / pending-ticket set doesn't already answer
* via {@code STOP}.
*/
public final class ReplyPushLoop {
@@ -60,14 +57,6 @@ public final class ReplyPushLoop {
/** Coalesced form, several uncollected tickets for the same lead. */
static final String TICKETS_NUDGE_FORMAT =
"%d tickets finished%s — run bridge_poll(ticket=...) for each to collect them: %s";
/** Singular form, one worker paused mid-turn in bridge_ask (CB-582) — names the answer call directly. */
static final String QUESTION_NUDGE_FORMAT =
"Worker %s asked a question (ticket %s) — answer it with bridge_send(turnId=\"%s\", "
+ "content=...) to resume its turn:\n%s";
/** Coalesced form, several open questions for the same lead. */
static final String QUESTIONS_NUDGE_FORMAT =
"%d workers are paused on a question — run bridge_poll(ticket=...) for each, then answer "
+ "with bridge_send(turnId=..., content=...): %s";
private final PrimaryRegistry primaryRegistry;
private final AgentControl agents;
@@ -86,14 +75,7 @@ public final class ReplyPushLoop {
private final ConcurrentHashMap<String, ReplyEntry> pendingReplies = new ConcurrentHashMap<>();
/** Tickets that have gone terminal but not yet been polled, keyed by ticket. */
private final ConcurrentHashMap<String, PendingTicket> pendingTickets = new ConcurrentHashMap<>();
/**
* Open {@code bridge_ask} questions not yet answered or lapsed, keyed by {@code turnId}
* (CB-582). A question's own nudge count is tracked the same per-item way as
* {@link #pendingTickets} (CB-598): a fresh question keeps its source eligible regardless of
* how depleted an older, still-open question's count is.
*/
private final ConcurrentHashMap<String, PendingQuestion> pendingQuestions = new ConcurrentHashMap<>();
/** CB-590: leads with an active combined reminder schedule (replies and/or tickets and/or questions). */
/** CB-590: leads with an active combined reminder schedule (replies and/or tickets). */
private final ConcurrentHashMap<String, Boolean> activeLeads = new ConcurrentHashMap<>();
public ReplyPushLoop(PrimaryRegistry primaryRegistry, AgentControl agents, ReplyInbox inbox,
@@ -172,26 +154,6 @@ public final class ReplyPushLoop {
.collect(Collectors.toUnmodifiableSet());
}
/**
* An open question awaiting the lead's answer: which ticket it belongs to, which worker asked,
* which lead to nudge, the question text, and how many nudges have named it so far (CB-598 —
* tracked per question, not per lead per source).
*/
private record PendingQuestion(String turnId, String ticket, String target, String lead,
String question, int nudgeCount) {
}
/** Questions still open for {@code lead}, snapshotted fresh for one tick. */
private List<PendingQuestion> pendingQuestionsFor(String lead) {
return pendingQuestions.values().stream().filter(q -> lead.equals(q.lead())).toList();
}
/** Question turnIds still open for {@code lead} — a plain snapshot for race comparison. */
private Set<String> pendingQuestionTurnIdsFor(String lead) {
return pendingQuestionsFor(lead).stream().map(PendingQuestion::turnId)
.collect(Collectors.toUnmodifiableSet());
}
/**
* The reply-source reminder count {@link #decide} should see for {@code lead} on this tick:
* the <em>minimum</em> nudge count among the reply targets currently pending for it (CB-598).
@@ -227,15 +189,6 @@ public final class ReplyPushLoop {
return min == Integer.MAX_VALUE ? 0 : min;
}
/** As {@link #minReplyNudgeCountFor}, for the question source (CB-582). */
private int minQuestionNudgeCountFor(String lead) {
int min = Integer.MAX_VALUE;
for (PendingQuestion q : pendingQuestionsFor(lead)) {
min = Math.min(min, q.nudgeCount());
}
return min == Integer.MAX_VALUE ? 0 : min;
}
/**
* Pure decision function: examine everything pending for {@code lead} — reply targets and
* tickets alike — and return what the loop should do.
@@ -260,29 +213,15 @@ public final class ReplyPushLoop {
* @return the action the caller should take
*/
Action decide(String lead, int replyReminderCount, int ticketReminderCount) {
return decide(lead, replyReminderCount, ticketReminderCount, minQuestionNudgeCountFor(lead));
}
/**
* As {@link #decide(String, int, int)}, with the question source (CB-582) folded in on the
* same footing as replies and tickets: its own eligibility (has open questions AND under its
* own {@link #maxReminders} budget) is enough on its own to {@link Action#INJECT}, exactly like
* the other two.
*
* @param questionReminderCount the lowest nudge count among questions open for this lead
*/
Action decide(String lead, int replyReminderCount, int ticketReminderCount, int questionReminderCount) {
boolean hasReplyWork = !pendingReplyTargetsFor(lead).isEmpty();
boolean hasTicketWork = !pendingTicketIdsFor(lead).isEmpty();
boolean hasQuestionWork = !pendingQuestionTurnIdsFor(lead).isEmpty();
if (!hasReplyWork && !hasTicketWork && !hasQuestionWork) {
if (!hasReplyWork && !hasTicketWork) {
log.debug("push: nothing pending for lead {}, stopping reminder", lead);
return Action.STOP;
}
boolean replyEligible = hasReplyWork && replyReminderCount < maxReminders;
boolean ticketEligible = hasTicketWork && ticketReminderCount < maxReminders;
boolean questionEligible = hasQuestionWork && questionReminderCount < maxReminders;
if (!replyEligible && !ticketEligible && !questionEligible) {
if (!replyEligible && !ticketEligible) {
log.debug("push: reminder cap ({}) reached for lead {} on every source with pending work, stopping",
maxReminders, lead);
countNudge("exhausted");
@@ -361,39 +300,6 @@ public final class ReplyPushLoop {
pendingTickets.remove(ticket);
}
/**
* Called when an async ticket's worker pauses mid-turn in {@code bridge_ask} (CB-582): the
* question is now visible via {@code bridge_poll} (Phase.ASKING), but the reverse-rendezvous
* window it opened with (~55s default, see {@code BridgeMcp}/{@code BridgedApp}) is far shorter
* than a lead's normal minutes-long poll cadence — exactly the gap this closes. Resolves the
* delegating lead the same way {@link #onTicketTerminal} does and coalesces onto the same
* per-lead schedule (CB-590).
*
* @param ticket the async ticket the question belongs to (for {@code bridge_poll})
* @param target the worker session that asked
* @param turnId correlation id the lead answers with ({@code bridge_send turnId=...})
* @param question the question text
*/
public void onQuestionOpened(String ticket, String target, String turnId, String question) {
var lead = primaryRegistry.nudgeTargetFor(target);
if (lead.isEmpty()) {
log.debug("push: no lead is known to be waiting on {}'s question (turnId {}), skipping nudge",
target, turnId);
return;
}
pendingQuestions.put(turnId, new PendingQuestion(turnId, ticket, target, lead.get(), question, 0));
startOrCoalesce(lead.get());
}
/**
* Called when a worker's {@code bridge_ask} resolves — answered or lapsed unanswered — so a
* scheduled tick never nudges about a question the lead already handled. A {@code turnId} that
* was never pending (never nudged, or already closed) is a no-op.
*/
public void questionClosed(String turnId) {
pendingQuestions.remove(turnId);
}
// --- the schedule ----------------------------------------------------------------------------
/** Start a reminder schedule for {@code lead}, or join the one already running. */
@@ -424,19 +330,17 @@ public final class ReplyPushLoop {
void tick(String lead) {
Set<String> repliesBefore = pendingReplyTargetsFor(lead);
Set<String> ticketsBefore = pendingTicketIdsFor(lead);
Set<String> questionsBefore = pendingQuestionTurnIdsFor(lead);
int replyReminderCount = minReplyNudgeCountFor(lead);
int ticketReminderCount = minTicketNudgeCountFor(lead);
int questionReminderCount = minQuestionNudgeCountFor(lead);
var action = decide(lead, replyReminderCount, ticketReminderCount, questionReminderCount);
var action = decide(lead, replyReminderCount, ticketReminderCount);
switch (action) {
case INJECT -> {
injectNudge(lead, replyReminderCount, ticketReminderCount, questionReminderCount);
injectNudge(lead, replyReminderCount, ticketReminderCount);
scheduleNext(lead);
}
// Re-check after the configured backoff; the lead may become injectable soon.
case WAIT_BUSY -> scheduleNext(lead);
case STOP -> stopOrRestart(lead, repliesBefore, ticketsBefore, questionsBefore);
case STOP -> stopOrRestart(lead, repliesBefore, ticketsBefore);
}
}
@@ -471,20 +375,9 @@ public final class ReplyPushLoop {
* side always wins; neither can miss the other, so this never loops on its own account.
*/
void stopOrRestart(String lead, Set<String> repliesBefore, Set<String> ticketsBefore) {
stopOrRestart(lead, repliesBefore, ticketsBefore, pendingQuestionTurnIdsFor(lead));
}
/**
* As {@link #stopOrRestart(String, Set, Set)}, with the question source's (CB-582) own "before"
* snapshot folded into the same race check: a question that raced in during the
* decision-to-release window reclaims the schedule slot exactly like a raced-in reply or ticket.
*/
void stopOrRestart(String lead, Set<String> repliesBefore, Set<String> ticketsBefore,
Set<String> questionsBefore) {
activeLeads.remove(lead);
boolean racedIn = pendingReplyTargetsFor(lead).stream().anyMatch(t -> !repliesBefore.contains(t))
|| pendingTicketIdsFor(lead).stream().anyMatch(t -> !ticketsBefore.contains(t))
|| pendingQuestionTurnIdsFor(lead).stream().anyMatch(t -> !questionsBefore.contains(t));
|| pendingTicketIdsFor(lead).stream().anyMatch(t -> !ticketsBefore.contains(t));
if (racedIn && activeLeads.putIfAbsent(lead, Boolean.TRUE) == null) {
log.debug("push: new work for lead {} raced the reminder loop's stop — restarting", lead);
scheduleNext(lead);
@@ -494,44 +387,36 @@ public final class ReplyPushLoop {
}
/** Send one combined nudge covering everything currently pending for {@code lead}. */
private void injectNudge(String lead, int replyReminderCount, int ticketReminderCount,
int questionReminderCount) {
// Re-read rather than threading it down from decide(): a reply can drain, a ticket be
// collected, or a question be answered (or another arrive), between the decision and the
// injection.
private void injectNudge(String lead, int replyReminderCount, int ticketReminderCount) {
// Re-read rather than threading it down from decide(): a reply can drain, or a ticket be
// collected (or another arrive), between the decision and the injection.
Set<String> replyTargets = pendingReplyTargetsFor(lead);
List<PendingTicket> tickets = pendingTicketsFor(lead);
List<PendingQuestion> questions = pendingQuestionsFor(lead);
if (replyTargets.isEmpty() && tickets.isEmpty() && questions.isEmpty()) {
if (replyTargets.isEmpty() && tickets.isEmpty()) {
log.debug("push: pending work for lead {} drained before the nudge could be sent", lead);
return;
}
String nudge = formatNudge(replyTargets, tickets, questions);
String nudge = formatNudge(replyTargets, tickets);
try {
agents.send(lead, nudge);
log.debug("push: nudge sent to lead {} (reply {}/{}, ticket {}/{}, question {}/{}; "
+ "{} reply target(s), {} ticket(s), {} question(s))",
log.debug("push: nudge sent to lead {} (reply {}/{}, ticket {}/{}; {} reply target(s), {} ticket(s))",
lead, replyReminderCount + 1, maxReminders, ticketReminderCount + 1, maxReminders,
questionReminderCount + 1, maxReminders,
replyTargets.size(), tickets.size(), questions.size());
replyTargets.size(), tickets.size());
countNudge("delivered");
} catch (RuntimeException e) {
log.warn("push: failed to nudge lead {} (reply {}/{}, ticket {}/{}, question {}/{}): {}",
lead, replyReminderCount + 1, maxReminders, ticketReminderCount + 1, maxReminders,
questionReminderCount + 1, maxReminders, e.toString());
log.warn("push: failed to nudge lead {} (reply {}/{}, ticket {}/{}): {}",
lead, replyReminderCount + 1, maxReminders, ticketReminderCount + 1, maxReminders, e.toString());
}
// Bump every item actually named in this nudge, not just whatever the shared source-level
// eligibility used to gate (CB-598) — each item's own count is what the next tick's
// minReplyNudgeCountFor / minTicketNudgeCountFor / minQuestionNudgeCountFor will read. An
// item already at or over the cap keeps riding along in the text (still pending, still
// named) but its extra bumps here are inert: decide() already treats it as ineligible once
// its count reaches maxReminders.
bumpNudgeCounts(replyTargets, tickets, questions);
// minReplyNudgeCountFor / minTicketNudgeCountFor will read. An item already at or over the
// cap keeps riding along in the text (still pending, still named) but its extra bumps here
// are inert: decide() already treats it as ineligible once its count reaches maxReminders.
bumpNudgeCounts(replyTargets, tickets);
}
/** Record that every one of these items was just named in a sent (or attempted) nudge. */
private void bumpNudgeCounts(Set<String> replyTargets, List<PendingTicket> tickets,
List<PendingQuestion> questions) {
private void bumpNudgeCounts(Set<String> replyTargets, List<PendingTicket> tickets) {
for (String target : replyTargets) {
pendingReplies.computeIfPresent(target, (t, e) -> new ReplyEntry(e.lead(), e.nudgeCount() + 1));
}
@@ -539,11 +424,6 @@ public final class ReplyPushLoop {
pendingTickets.computeIfPresent(ticket.ticket(),
(id, e) -> new PendingTicket(e.ticket(), e.lead(), e.failed(), e.nudgeCount() + 1));
}
for (PendingQuestion question : questions) {
pendingQuestions.computeIfPresent(question.turnId(), (id, e) ->
new PendingQuestion(e.turnId(), e.ticket(), e.target(), e.lead(), e.question(),
e.nudgeCount() + 1));
}
}
/** Schedule the next tick on the scheduler thread pool. */
@@ -555,8 +435,7 @@ public final class ReplyPushLoop {
// --- nudge formatting ------------------------------------------------------------------------
/** Render everything pending for one lead as a single nudge line. */
private static String formatNudge(Set<String> replyTargets, List<PendingTicket> tickets,
List<PendingQuestion> questions) {
private static String formatNudge(Set<String> replyTargets, List<PendingTicket> tickets) {
List<String> parts = new ArrayList<>();
if (!replyTargets.isEmpty()) {
parts.add(formatRepliesNudge(replyTargets));
@@ -564,9 +443,6 @@ public final class ReplyPushLoop {
if (!tickets.isEmpty()) {
parts.add(formatTicketsNudge(tickets));
}
if (!questions.isEmpty()) {
parts.add(formatQuestionsNudge(questions));
}
return String.join(" | ", parts);
}
@@ -594,18 +470,6 @@ public final class ReplyPushLoop {
return TICKETS_NUDGE_FORMAT.formatted(pending.size(), failedNote, ids);
}
/** Render one or several open questions (CB-582). */
private static String formatQuestionsNudge(List<PendingQuestion> pending) {
if (pending.size() == 1) {
PendingQuestion q = pending.get(0);
return QUESTION_NUDGE_FORMAT.formatted(q.target(), q.ticket(), q.turnId(), q.question());
}
String ids = pending.stream()
.map(q -> q.ticket() + " (turnId=" + q.turnId() + ")")
.collect(Collectors.joining(", "));
return QUESTIONS_NUDGE_FORMAT.formatted(pending.size(), ids);
}
// --- lifecycle -----------------------------------------------------------------------------
/**
@@ -613,8 +477,8 @@ public final class ReplyPushLoop {
* uses this to stand aside: while the push loop is actively nudging a lead, a concurrent
* heartbeat injection would start a second competing turn in the same pane — racing loops
* multiply turns and context burn. "Active" means a schedule exists in {@link #activeLeads},
* which now covers reply-queued (CB-307), ticket-terminal (CB-588), and question-open (CB-582)
* work (CB-590) — bounded by what has been triggered, not by any persistent state.
* which now covers both reply-queued (CB-307) and ticket-terminal (CB-588) work (CB-590) —
* bounded by what has been triggered, not by any persistent state.
*/
public boolean isActive() {
return !activeLeads.isEmpty();
@@ -626,7 +490,6 @@ public final class ReplyPushLoop {
activeLeads.clear();
pendingReplies.clear();
pendingTickets.clear();
pendingQuestions.clear();
}
/** @see #stop() */
@@ -509,20 +509,10 @@ public final class BridgedApp {
return;
}
try {
Map<String, Object> body = new LinkedHashMap<>();
body.put("sessionId", id);
body.put("status", messages.status(id).name().toLowerCase());
body.put("ready", presence.isPresent(id));
// CB-582: a worker paused mid-turn in an async bridge_ask is otherwise invisible to a
// status poll — surface the open question and how to answer it, same as bridge_poll's
// Phase.ASKING view.
MessageService.PendingAsk ask = messages.pendingAsk(id);
if (ask != null) {
body.put("question", ask.question());
body.put("turnId", ask.turnId());
body.put("ticket", ask.ticket());
}
ctx.status(200).json(body);
ctx.status(200).json(Map.of(
"sessionId", id,
"status", messages.status(id).name().toLowerCase(),
"ready", presence.isPresent(id)));
} catch (HerdrException e) {
herdrError(ctx, e);
}
@@ -548,11 +538,6 @@ public final class BridgedApp {
if (v.detail() != null) {
body.put("detail", v.detail());
}
// CB-582: Phase.ASKING carries the question in v.reply() (handled above) and its answer-
// correlation id here — a REST caller polling this ticket otherwise has no way to answer it.
if (v.turnId() != null) {
body.put("turnId", v.turnId());
}
ctx.status(200).json(body);
}
@@ -293,8 +293,10 @@ public final class SessionManager implements TurnListener {
// blocked caller fails fast with a real reason instead of sitting on a rendezvous
// nothing will ever resolve. CB-578 stage C: carry the worktree/branch/snapshot ref
// too, so a failed ticket's detail can point a lead at the same tree to re-dispatch.
// CB-584 (issue #65 criterion 5): carry agentSessionId alongside them, so a lead can
// also resume the member's conversation, not just re-dispatch onto its files.
notifyReleased(new ReleaseDetail(removed.terminalId(), removed.worktree(),
removed.branch(), snapshotRef));
removed.branch(), snapshotRef, removed.agentSessionId()));
}
}
// CB-581: the pane must always stop, even if the dirty check above threw. A session removed
@@ -348,7 +350,8 @@ public final class SessionManager implements TurnListener {
* are {@code null} for a shared-tree session; {@code snapshotRef} is {@code null} unless this
* release snapshotted a dirty worktree into {@code refs/wip/<branch>}.
*/
public record ReleaseDetail(String terminalId, String worktreePath, String branch, String snapshotRef) {
public record ReleaseDetail(String terminalId, String worktreePath, String branch, String snapshotRef,
String agentSessionId) {
}
/**
@@ -1040,6 +1040,44 @@ class BridgedConfigTest {
"an opencode worker with no argv defaults to the opencode binary, never claude");
}
/**
* CB-604: an unrecognized {@code kind:} used to silently fall into the claude-code bucket — not
* matching {@code "opencode"} was the only check. With {@code argv:} also unset that meant the
* daemon tried to launch a program literally named after the typo.
*/
@Test
void unknownKindIsRefusedAtLoadNamingTheValueAndTheAcceptedSet(@TempDir Path dir) throws Exception {
Path f = dir.resolve("kind-typo.yaml");
Files.writeString(f, """
profiles:
gemini:
kind: opencod
model: google/gemini-2.5-pro
""");
IllegalStateException e = assertThrows(IllegalStateException.class, () -> BridgedConfig.load(f));
assertTrue(e.getMessage().contains("gemini"), "error names the profile: " + e.getMessage());
assertTrue(e.getMessage().contains("opencod"), "error names the bad value: " + e.getMessage());
assertTrue(e.getMessage().contains("claude-code") && e.getMessage().contains("opencode"),
"error names the accepted set: " + e.getMessage());
}
@Test
void blankKindStillDefaultsToClaudeCode(@TempDir Path dir) throws Exception {
Path f = dir.resolve("kind-blank.yaml");
Files.writeString(f, """
profiles:
gx10:
baseUrl: http://gx10.gw:8000
kind: ""
argv: ["ccs", "gx10"]
""");
BridgedConfig cfg = BridgedConfig.load(f);
assertEquals(BridgedConfig.Profile.KIND_CLAUDE_CODE, cfg.profiles().get("gx10").kind(),
"a blank kind: is documented to behave exactly like an absent one");
}
@Test
void authDefaultsToLoopbackTrustSoExistingConfigsBehaveAsBefore(@TempDir Path dir) throws Exception {
Path f = dir.resolve("no-auth-block.yaml");
@@ -773,52 +773,6 @@ class BridgeMcpTest {
assertEquals("blocked", textOf(res));
}
/**
* CB-582: a lead polling {@code bridge_status} on its normal cadence — not {@code bridge_poll}
* — must also see a worker's open async {@code bridge_ask} question, since the reverse-rendezvous
* window it opened with is far shorter than that cadence.
*/
@Test
void statusReportsAnOpenQuestionWhenTheWorkerIsMidAsk() throws Exception {
String ticket = messages.sendAsync(T, "task that asks");
long deadline = System.currentTimeMillis() + 3000;
while (!rendezvous.isWaiting(T) && System.currentTimeMillis() < deadline) {
Thread.sleep(5);
}
assertTrue(rendezvous.isWaiting(T), "sendAsync should have opened its rendezvous waiter");
CompletableFuture<MessageService.AskResult> ask =
CompletableFuture.supplyAsync(() -> messages.ask(T, "which config file?", 5000));
MessageService.TaskView asking;
deadline = System.currentTimeMillis() + 3000;
do {
asking = messages.poll(ticket);
Thread.sleep(5);
} while (asking.phase() != MessageService.Phase.ASKING && System.currentTimeMillis() < deadline);
assertEquals(MessageService.Phase.ASKING, asking.phase());
McpSchema.CallToolResult res = BridgeMcp.status(messages, T);
assertNotEquals(Boolean.TRUE, res.isError());
String out = textOf(res);
assertTrue(out.startsWith("idle"), "the live status must still lead the text: " + out);
assertTrue(out.contains("which config file?"), "the question text must be shown: " + out);
assertTrue(out.contains("turnId=\"" + asking.turnId() + "\""), "the turnId must be shown: " + out);
assertTrue(out.contains("(ticket " + ticket + ")"), "the ticket must be shown: " + out);
// Clean up the still-open ask so the background thread does not linger past the test.
String turnId = asking.turnId();
CompletableFuture<MessageService.Reply> answer = CompletableFuture.supplyAsync(
() -> messages.answer(turnId, "config.yaml", 5000));
assertEquals("config.yaml", ask.get(5, TimeUnit.SECONDS).answer());
deadline = System.currentTimeMillis() + 3000;
while (!rendezvous.isWaiting(T) && System.currentTimeMillis() < deadline) {
Thread.sleep(5);
}
assertTrue(rendezvous.resolve(T, "done"));
answer.get(5, TimeUnit.SECONDS);
}
// --- bridge_whoami: the caller's own identity, so an agent never has to guess its role -------
@Test
@@ -714,47 +714,6 @@ class MessageServiceTest {
assertEquals(MessageService.Outcome.REPLIED, answer.get(5, TimeUnit.SECONDS).outcome());
}
// --- CB-582: bridge_status pendingAsk() ------------------------------------------------------
@Test
void pendingAskReturnsNullWhenNoQuestionIsOpen() throws Exception {
assertNull(messages.pendingAsk(T), "no async ticket at all -> no pending ask");
String ticket = messages.sendAsync(T, "long task");
awaitWaiting();
assertNull(messages.pendingAsk(T), "a plain pending delegation is not a question");
injectDelivery();
assertTrue(rendezvous.resolve(T, "done"));
awaitTicketPhase(ticket, MessageService.Phase.DONE);
assertNull(messages.pendingAsk(T), "a finished ticket carries no open question either");
}
@Test
void pendingAskReturnsTheOpenQuestionForAnAsyncTicket() throws Exception {
String ticket = messages.sendAsync(T, "task that asks");
awaitWaiting();
injectDelivery();
CompletableFuture<MessageService.AskResult> ask =
CompletableFuture.supplyAsync(() -> messages.ask(T, "which config file?", 5000));
MessageService.TaskView asking = awaitTicketPhase(ticket, MessageService.Phase.ASKING);
MessageService.PendingAsk pending = messages.pendingAsk(T);
assertNotNull(pending, "bridge_status should see the open question");
assertEquals(ticket, pending.ticket());
assertEquals("which config file?", pending.question());
assertEquals(asking.turnId(), pending.turnId());
CompletableFuture<MessageService.Reply> answer = CompletableFuture.supplyAsync(
() -> messages.answer(asking.turnId(), "config.yaml", 5000));
assertEquals("config.yaml", ask.get(5, TimeUnit.SECONDS).answer());
assertNull(messages.pendingAsk(T), "an answered question is no longer pending");
awaitWaiting();
assertTrue(rendezvous.resolve(T, "done"));
assertEquals(MessageService.Outcome.REPLIED, answer.get(5, TimeUnit.SECONDS).outcome());
}
// --- CB-588: async ticket terminal nudges ---------------------------------------------------
//
// MessageService.reply's rendezvous fast path is exactly what an async ticket always takes
@@ -874,71 +833,6 @@ class MessageServiceTest {
}
}
// --- CB-582: bridge_ask question-open nudges --------------------------------------------------
@Test
void anAsyncTicketThatPausesOnAQuestionNudgesTheLeadWithNoPriorPollCall() throws Exception {
try (var wiring = wireWithPushLoop(5, 50)) {
String ticket = wiring.service().sendAsync(T, "task that asks");
awaitWaiting();
injectDelivery();
CompletableFuture<MessageService.AskResult> ask = CompletableFuture.supplyAsync(
() -> wiring.service().ask(T, "which config file?", 5000));
MessageService.TaskView asking = awaitTicketPhaseOn(wiring.service(), ticket, MessageService.Phase.ASKING);
awaitNudge(wiring.leadHerdr());
String nudge = wiring.leadHerdr().lastCall("agent.prompt").params().toString();
assertTrue(nudge.contains(ticket), "the nudge should name the ticket: " + nudge);
assertTrue(nudge.contains(asking.turnId()), "the nudge should name the turnId: " + nudge);
assertTrue(nudge.contains("bridge_send(turnId="),
"the nudge should name the exact answer call: " + nudge);
assertTrue(nudge.contains("which config file?"), "the nudge should include the question: " + nudge);
// Clean up the still-open ask so the background thread does not linger past the test.
CompletableFuture<MessageService.Reply> answer = CompletableFuture.supplyAsync(
() -> wiring.service().answer(asking.turnId(), "config.yaml", 5000));
assertEquals("config.yaml", ask.get(5, TimeUnit.SECONDS).answer());
awaitWaiting();
assertTrue(rendezvous.resolve(T, "done"));
answer.get(5, TimeUnit.SECONDS);
}
}
@Test
void answeringAQuestionStopsFurtherNudgesAboutIt() throws Exception {
try (var wiring = wireWithPushLoop(5, 50)) {
String ticket = wiring.service().sendAsync(T, "task that asks");
awaitWaiting();
injectDelivery();
CompletableFuture<MessageService.AskResult> ask = CompletableFuture.supplyAsync(
() -> wiring.service().ask(T, "which config file?", 5000));
MessageService.TaskView asking = awaitTicketPhaseOn(wiring.service(), ticket, MessageService.Phase.ASKING);
awaitNudge(wiring.leadHerdr());
long callsBeforeAnswer = wiring.leadHerdr().calls.stream()
.filter(c -> c.method().equals("agent.prompt")).count();
CompletableFuture<MessageService.Reply> answer = CompletableFuture.supplyAsync(
() -> wiring.service().answer(asking.turnId(), "config.yaml", 5000));
assertEquals("config.yaml", ask.get(5, TimeUnit.SECONDS).answer());
awaitWaiting();
assertTrue(rendezvous.resolve(T, "done"));
answer.get(5, TimeUnit.SECONDS);
// Let several more ticks (and the ticket's own now-legitimate terminal nudge) fire —
// none of them may still name the question's turnId, which is closed.
Thread.sleep(300);
boolean anyNamesClosedQuestion = wiring.leadHerdr().calls.stream()
.filter(c -> c.method().equals("agent.prompt"))
.skip(callsBeforeAnswer)
.anyMatch(c -> c.params().toString().contains(asking.turnId()));
assertFalse(anyNamesClosedQuestion,
"no nudge sent after the answer may still name the now-closed turnId " + asking.turnId());
}
}
@Test
void aFleetWithNoPushLoopConfiguredBehavesExactlyAsToday() throws Exception {
// `messages` (the shared field) uses the no-pushLoop constructor — poll() must not throw,
@@ -657,132 +657,6 @@ class ReplyPushLoopTest {
assertTrue(loop.isActive(), "the schedule must stay active after nudging the fresh ticket");
}
// --- CB-582: bridge_ask question-open nudges -------------------------------------------------
@Test
void onQuestionOpenedWithNoKnownLeadNeverStartsASchedule() throws Exception {
var rec = recordingClient();
agents = new AgentControl(rec);
var loop = new ReplyPushLoop(new PrimaryRegistry(null), agents, inbox, scheduler, 5, 50);
loop.onQuestionOpened("task-1", WORKER, "term_worker#1", "which config?");
assertFalse(loop.isActive(), "no lead known means nothing to nudge yet");
Thread.sleep(150);
assertEquals(0, rec.sendCount(), "must not nudge when no lead is known to be waiting");
}
@Test
void decideQuestionsWithNothingPendingIsStop() {
agents = agentWithStatus("idle");
assertEquals(ReplyPushLoop.Action.STOP, loop().decide(PRIMARY, 0, 0, 0));
}
@Test
void decideQuestionsAtCapIsStop() {
agents = agentWithStatus("idle");
var loop = loop(2, 100_000);
loop.onQuestionOpened("task-1", WORKER, "term_worker#1", "which config?");
assertEquals(ReplyPushLoop.Action.STOP, loop.decide(PRIMARY, 0, 0, 2));
}
@Test
void decideQuestionsUnderCapWithInjectableLeadIsInject() {
agents = agentWithStatus("idle");
var loop = loop(5, 100_000);
loop.onQuestionOpened("task-1", WORKER, "term_worker#1", "which config?");
assertEquals(ReplyPushLoop.Action.INJECT, loop.decide(PRIMARY, 0, 0, 0));
}
@Test
void onQuestionOpenedCausesExactlyOneNudgeNamingTheTicketAndTurnId() throws Exception {
var rec = recordingClient();
agents = new AgentControl(rec);
loop(1, 50).onQuestionOpened("task-1", WORKER, "term_worker#1", "which config file?");
assertTrue(rec.sendLatch.await(3, TimeUnit.SECONDS), "one question nudge should have been sent");
assertEquals(1, rec.sendCount());
String nudge = rec.sentParams().getFirst().getValue().toString();
assertTrue(nudge.contains("task-1"), "nudge should name the ticket: " + nudge);
assertTrue(nudge.contains("term_worker#1"), "nudge should name the turnId: " + nudge);
assertTrue(nudge.contains("bridge_send(turnId="), "nudge should name the exact answer call: " + nudge);
assertTrue(nudge.contains("which config file?"), "nudge should include the question text: " + nudge);
}
@Test
void questionClosedPreventsFurtherNudging() throws Exception {
var rec = recordingClient();
agents = new AgentControl(rec);
var loop = loop(1, 100);
loop.onQuestionOpened("task-1", WORKER, "term_worker#1", "which config?");
loop.questionClosed("term_worker#1"); // answered/lapsed before the first tick fired
Thread.sleep(300); // let the scheduled tick run
assertEquals(0, rec.sendCount(), "an already-closed question must never be nudged");
}
@Test
void questionNudgesSendUpToCapThenStop() throws Exception {
int cap = 2;
var rec = recordingClient();
agents = new AgentControl(rec);
rec.sendLatch = new CountDownLatch(cap);
loop(cap, 50).onQuestionOpened("task-1", WORKER, "term_worker#1", "which config?");
assertTrue(rec.sendLatch.await(5, TimeUnit.SECONDS), cap + " question nudges should have fired");
Thread.sleep(300);
assertEquals(cap, rec.sendCount(), "exactly " + cap + " question nudges (cap=" + cap + ")");
}
@Test
void questionAndTicketForTheSameLeadCoalesceIntoOneSend() throws Exception {
var rec = recordingClient();
agents = new AgentControl(rec);
var loop = loop(1, 300); // backoff wide enough that both entry points land before the first tick
loop.onTicketTerminal("task-1", WORKER, false);
loop.onQuestionOpened("task-2", WORKER, "term_worker#1", "which config?");
assertTrue(rec.sendLatch.await(3, TimeUnit.SECONDS), "one combined nudge should have been sent");
Thread.sleep(300);
assertEquals(1, rec.sendCount(),
"a ticket and a question for the same lead must coalesce onto ONE schedule");
String nudge = rec.sentParams().getFirst().getValue().toString();
assertTrue(nudge.contains("task-1"), "the combined nudge must still mention the ticket: " + nudge);
assertTrue(nudge.contains("term_worker#1"), "the combined nudge must still mention the question: " + nudge);
}
@Test
void oneExhaustedQuestionSourceDoesNotBlockANudgeForTheOtherSources() {
// Mirrors oneExhaustedSourceDoesNotBlockANudgeForTheOtherSource for the question source:
// the question source is at its cap (2/2), but the ticket source has never been nudged
// (0/2) — decide() must still INJECT so the ticket is not stranded.
agents = agentWithStatus("idle");
var loop = loop(2, 100_000);
loop.onQuestionOpened("task-1", WORKER, "term_worker#1", "which config?");
loop.onTicketTerminal("task-2", WORKER, false);
assertEquals(ReplyPushLoop.Action.INJECT, loop.decide(PRIMARY, 0, 0, 2),
"the question source is exhausted (2/2), but the ticket source has never been "
+ "nudged (0/2) — the lead must still be injected so the ticket is not lost");
}
@Test
void questionNudgeFormatIsCorrect() {
String single = ReplyPushLoop.QUESTION_NUDGE_FORMAT.formatted(
WORKER, "task-1", "term_worker#1", "which config?");
assertTrue(single.contains("Worker term_worker"));
assertTrue(single.contains("bridge_send(turnId=\"term_worker#1\""));
assertTrue(single.contains("which config?"));
String multi = ReplyPushLoop.QUESTIONS_NUDGE_FORMAT.formatted(2, "task-1 (turnId=t1), task-2 (turnId=t2)");
assertTrue(multi.contains("2 workers"));
assertTrue(multi.contains("bridge_poll(ticket=...)"));
}
// --- metrics (CB-512) ----------------------------------------------------------------------
@Test
@@ -484,67 +484,6 @@ class BridgedAppTest {
assertTrue(mapper.readTree(req(port, "GET", "/sessions/term_a/status").body()).get("ready").asBoolean());
}
/**
* CB-582: a lead polling {@code GET /sessions/{id}/status} on its normal cadence — not the
* ticket-scoped {@code /tasks/{ticket}} — must also see a worker's open async {@code bridge_ask}
* question, since the reverse-rendezvous window it opened with is far shorter than that cadence.
*/
@Test
void sessionStatusReportsAnOpenQuestionWhenTheWorkerIsMidAsk() throws Exception {
FakeHerdr herdr = new FakeHerdr().agentStatus("idle"); // poller delivers the injection
int port = start(herdr, "http://gx00.gw:8000", Set.of("gx00.gw"));
HttpResponse<String> accepted = postMessage(port, "{\"content\":\"do it\",\"wait\":false}");
assertEquals(202, accepted.statusCode());
String ticket = mapper.readTree(accepted.body()).get("ticket").asText();
Thread.sleep(200); // let the background async send open its rendezvous waiter
var ask = java.util.concurrent.CompletableFuture.supplyAsync(() -> {
try {
return postJson(port, "/sessions/term_a/ask",
"{\"question\":\"which config file?\",\"timeoutMs\":5000}");
} catch (Exception e) {
throw new RuntimeException(e);
}
});
JsonNode task;
long deadline = System.currentTimeMillis() + 3000;
do {
task = mapper.readTree(req(port, "GET", "/tasks/" + ticket).body());
if ("asking".equals(task.path("phase").asText())) break;
//noinspection BusyWait
Thread.sleep(10);
} while (System.currentTimeMillis() < deadline);
assertEquals("asking", task.get("phase").asText());
String turnId = task.get("turnId").asText();
HttpResponse<String> status = req(port, "GET", "/sessions/term_a/status");
assertEquals(200, status.statusCode());
JsonNode body = mapper.readTree(status.body());
assertEquals("idle", body.get("status").asText(), "the live status must still be reported");
assertEquals("which config file?", body.get("question").asText());
assertEquals(turnId, body.get("turnId").asText());
assertEquals(ticket, body.get("ticket").asText());
// Answer it — via the same /message route bridge_send uses, keyed by turnId — so the
// background ask thread does not linger past the test.
var answer = java.util.concurrent.CompletableFuture.supplyAsync(() -> {
try {
return postJson(port, "/sessions/term_a/message",
"{\"content\":\"config.yaml\",\"turnId\":\"" + turnId + "\",\"timeoutMs\":4000}");
} catch (Exception e) {
throw new RuntimeException(e);
}
});
HttpResponse<String> askResponse = ask.get(6, java.util.concurrent.TimeUnit.SECONDS);
assertEquals(200, askResponse.statusCode());
assertEquals("config.yaml", mapper.readTree(askResponse.body()).get("answer").asText());
postJson(port, "/sessions/term_a/reply", "{\"content\":\"done\"}");
answer.get(6, java.util.concurrent.TimeUnit.SECONDS);
}
@Test
void stopWorkerInPanePlacementClosesOnlyThePane() throws Exception {
FakeHerdr herdr = new FakeHerdr();
@@ -422,6 +422,28 @@ class WorktreeSessionManagerTest {
"acceptance criterion 6: a failed ticket's detail must carry the snapshot ref");
}
@Test
void releaseNotifiesTheListenerWithTheAgentSessionId() {
// CB-584 (issue #65 criterion 5): a failed ticket's detail must also name the agent
// session, alongside worktree/branch/snapshot, so a lead can resume the conversation
// rather than only re-dispatch a fresh member onto the same files.
FakeHerdr herdr = new FakeHerdr();
FakeWorktrees worktrees = new FakeWorktrees().withRepoRoot("/repo").withPrefix("/wt")
.withDirty(true);
SessionManager sessions = new SessionManager(workerService(herdr), worktrees);
java.util.List<SessionManager.ReleaseDetail> released = new java.util.concurrent.CopyOnWriteArrayList<>();
sessions.onRelease(released::add);
MemberSession s = sessions.acquire("ltms-local", MemberRole.DEV, null, "/caller/proj", null,
new WorktreeRequest("cb-584-e", null), "cb-584-session", null);
assertNotNull(s.agentSessionId(), "a named session must mint an agent session id to assert on");
sessions.release(s.paneId());
assertEquals(1, released.size());
SessionManager.ReleaseDetail detail = released.getFirst();
assertEquals(s.agentSessionId(), detail.agentSessionId());
}
@Test
void aFailingSnapshotStillPreservesTheWorktreeStopsThePaneAndNotifies() {
FakeHerdr herdr = new FakeHerdr();