agent-working-memory 0.7.16 → 0.8.5

package/dist/coordination/circuit-breaker.d.ts

@@ -0,0 +1,24 @@
+/**
+ * Per-worker CircuitBreaker for the coordination control layer.
+ * Prevents chronically-stale workers from poisoning the assignment queue.
+ * Part of AWM 0.8.1 — additive, no breaking changes.
+ *
+ * States:
+ *   closed    — normal operation
+ *   open      — worker blocked after FAILURE_THRESHOLD consecutive failures
+ *   half_open — probe window (30s after open), allows one assignment attempt
+ */
+import type Database from 'better-sqlite3';
+export type CircuitState = 'closed' | 'open' | 'half_open';
+/** Record a worker failure. Opens the circuit when consecutive failures hit the threshold. */
+export declare function recordFailure(db: Database.Database, agentId: string): void;
+/** Record a worker success. Resets to closed regardless of prior state. */
+export declare function recordSuccess(db: Database.Database, agentId: string): void;
+/**
+ * Get current circuit state for a worker.
+ * If the circuit has been open for >30s, auto-transitions to half_open.
+ */
+export declare function getState(db: Database.Database, agentId: string): CircuitState;
+/** Returns true when the worker is eligible to receive an assignment. */
+export declare function isAvailable(db: Database.Database, agentId: string): boolean;
+//# sourceMappingURL=circuit-breaker.d.ts.map

package/dist/coordination/circuit-breaker.d.ts.map

@@ -0,0 +1 @@
+{"version":3,"file":"circuit-breaker.d.ts","sourceRoot":"","sources":["../../src/coordination/circuit-breaker.ts"],"names":[],"mappings":"AAEA;;;;;;;;;GASG;AAEH,OAAO,KAAK,QAAQ,MAAM,gBAAgB,CAAC;AAE3C,MAAM,MAAM,YAAY,GAAG,QAAQ,GAAG,MAAM,GAAG,WAAW,CAAC;AAK3D,8FAA8F;AAC9F,wBAAgB,aAAa,CAAC,EAAE,EAAE,QAAQ,CAAC,QAAQ,EAAE,OAAO,EAAE,MAAM,GAAG,IAAI,CAiB1E;AAED,2EAA2E;AAC3E,wBAAgB,aAAa,CAAC,EAAE,EAAE,QAAQ,CAAC,QAAQ,EAAE,OAAO,EAAE,MAAM,GAAG,IAAI,CAU1E;AAED;;;GAGG;AACH,wBAAgB,QAAQ,CAAC,EAAE,EAAE,QAAQ,CAAC,QAAQ,EAAE,OAAO,EAAE,MAAM,GAAG,YAAY,CAoB7E;AAED,yEAAyE;AACzE,wBAAgB,WAAW,CAAC,EAAE,EAAE,QAAQ,CAAC,QAAQ,EAAE,OAAO,EAAE,MAAM,GAAG,OAAO,CAE3E"}

package/dist/coordination/circuit-breaker.js

@@ -0,0 +1,70 @@
+// Copyright 2026 Robert Winter / Complete Ideas
+// SPDX-License-Identifier: Apache-2.0
+/**
+ * Per-worker CircuitBreaker for the coordination control layer.
+ * Prevents chronically-stale workers from poisoning the assignment queue.
+ * Part of AWM 0.8.1 — additive, no breaking changes.
+ *
+ * States:
+ *   closed    — normal operation
+ *   open      — worker blocked after FAILURE_THRESHOLD consecutive failures
+ *   half_open — probe window (30s after open), allows one assignment attempt
+ */
+const FAILURE_THRESHOLD = 5;
+const HALF_OPEN_DELAY_MS = 30_000;
+/** Record a worker failure. Opens the circuit when consecutive failures hit the threshold. */
+export function recordFailure(db, agentId) {
+    db.prepare(`
+    INSERT INTO coord_circuit_state (agent_id, consecutive_failures, last_transition_at)
+    VALUES (?, 1, datetime('now'))
+    ON CONFLICT(agent_id) DO UPDATE SET
+      consecutive_failures = consecutive_failures + 1,
+      state = CASE
+        WHEN consecutive_failures + 1 >= ${FAILURE_THRESHOLD} THEN 'open'
+        ELSE state
+      END,
+      opened_at = CASE
+        WHEN consecutive_failures + 1 >= ${FAILURE_THRESHOLD} AND (state != 'open' OR opened_at IS NULL)
+        THEN datetime('now')
+        ELSE opened_at
+      END,
+      last_transition_at = datetime('now')
+  `).run(agentId);
+}
+/** Record a worker success. Resets to closed regardless of prior state. */
+export function recordSuccess(db, agentId) {
+    db.prepare(`
+    INSERT INTO coord_circuit_state (agent_id, state, consecutive_failures, last_transition_at)
+    VALUES (?, 'closed', 0, datetime('now'))
+    ON CONFLICT(agent_id) DO UPDATE SET
+      state = 'closed',
+      consecutive_failures = 0,
+      opened_at = NULL,
+      last_transition_at = datetime('now')
+  `).run(agentId);
+}
+/**
+ * Get current circuit state for a worker.
+ * If the circuit has been open for >30s, auto-transitions to half_open.
+ */
+export function getState(db, agentId) {
+    const row = db.prepare(`SELECT state, opened_at FROM coord_circuit_state WHERE agent_id = ?`).get(agentId);
+    if (!row || row.state === 'closed')
+        return 'closed';
+    if (row.state === 'half_open')
+        return 'half_open';
+    // open — check if half-open window has elapsed
+    if (row.state === 'open' && row.opened_at) {
+        const openedAt = new Date(row.opened_at.endsWith('Z') ? row.opened_at : row.opened_at + 'Z').getTime();
+        if (Date.now() - openedAt > HALF_OPEN_DELAY_MS) {
+            db.prepare(`UPDATE coord_circuit_state SET state = 'half_open', last_transition_at = datetime('now') WHERE agent_id = ?`).run(agentId);
+            return 'half_open';
+        }
+    }
+    return 'open';
+}
+/** Returns true when the worker is eligible to receive an assignment. */
+export function isAvailable(db, agentId) {
+    return getState(db, agentId) !== 'open';
+}
+//# sourceMappingURL=circuit-breaker.js.map

package/dist/coordination/circuit-breaker.js.map

@@ -0,0 +1 @@
+{"version":3,"file":"circuit-breaker.js","sourceRoot":"","sources":["../../src/coordination/circuit-breaker.ts"],"names":[],"mappings":"AAAA,gDAAgD;AAChD,sCAAsC;AACtC;;;;;;;;;GASG;AAMH,MAAM,iBAAiB,GAAG,CAAC,CAAC;AAC5B,MAAM,kBAAkB,GAAG,MAAM,CAAC;AAElC,8FAA8F;AAC9F,MAAM,UAAU,aAAa,CAAC,EAAqB,EAAE,OAAe;IAClE,EAAE,CAAC,OAAO,CAAC;;;;;;2CAM8B,iBAAiB;;;;2CAIjB,iBAAiB;;;;;GAKzD,CAAC,CAAC,GAAG,CAAC,OAAO,CAAC,CAAC;AAClB,CAAC;AAED,2EAA2E;AAC3E,MAAM,UAAU,aAAa,CAAC,EAAqB,EAAE,OAAe;IAClE,EAAE,CAAC,OAAO,CAAC;;;;;;;;GAQV,CAAC,CAAC,GAAG,CAAC,OAAO,CAAC,CAAC;AAClB,CAAC;AAED;;;GAGG;AACH,MAAM,UAAU,QAAQ,CAAC,EAAqB,EAAE,OAAe;IAC7D,MAAM,GAAG,GAAG,EAAE,CAAC,OAAO,CACpB,qEAAqE,CACtE,CAAC,GAAG,CAAC,OAAO,CAA4D,CAAC;IAE1E,IAAI,CAAC,GAAG,IAAI,GAAG,CAAC,KAAK,KAAK,QAAQ;QAAE,OAAO,QAAQ,CAAC;IACpD,IAAI,GAAG,CAAC,KAAK,KAAK,WAAW;QAAE,OAAO,WAAW,CAAC;IAElD,+CAA+C;IAC/C,IAAI,GAAG,CAAC,KAAK,KAAK,MAAM,IAAI,GAAG,CAAC,SAAS,EAAE,CAAC;QAC1C,MAAM,QAAQ,GAAG,IAAI,IAAI,CAAC,GAAG,CAAC,SAAS,CAAC,QAAQ,CAAC,GAAG,CAAC,CAAC,CAAC,CAAC,GAAG,CAAC,SAAS,CAAC,CAAC,CAAC,GAAG,CAAC,SAAS,GAAG,GAAG,CAAC,CAAC,OAAO,EAAE,CAAC;QACvG,IAAI,IAAI,CAAC,GAAG,EAAE,GAAG,QAAQ,GAAG,kBAAkB,EAAE,CAAC;YAC/C,EAAE,CAAC,OAAO,CACR,6GAA6G,CAC9G,CAAC,GAAG,CAAC,OAAO,CAAC,CAAC;YACf,OAAO,WAAW,CAAC;QACrB,CAAC;IACH,CAAC;IAED,OAAO,MAAM,CAAC;AAChB,CAAC;AAED,yEAAyE;AACzE,MAAM,UAAU,WAAW,CAAC,EAAqB,EAAE,OAAe;IAChE,OAAO,QAAQ,CAAC,EAAE,EAAE,OAAO,CAAC,KAAK,MAAM,CAAC;AAC1C,CAAC"}

package/dist/coordination/failure-modes.d.ts

@@ -0,0 +1,21 @@
+/**
+ * FailureMode taxonomy and mutation-hint map for the coordination control layer.
+ * Part of AWM 0.8.1 — additive, no breaking changes.
+ */
+export declare enum FailureMode {
+    AGENT_STALE = "agent_stale",
+    TIMEOUT = "timeout",
+    OUTPUT_INVALID = "output_invalid",
+    TEST_FAIL = "test_fail",
+    LINT_FAIL = "lint_fail",
+    MERGE_CONFLICT = "merge_conflict",
+    UNKNOWN = "unknown"
+}
+/** Classify a failure result string into one of the known modes. */
+export declare function classifyFailure(result: string | null): FailureMode;
+/**
+ * Corrective guidance injected into the task description on retry.
+ * Each hint is written in the vocabulary the next worker will read.
+ */
+export declare const MUTATION_HINTS: Record<FailureMode, string>;
+//# sourceMappingURL=failure-modes.d.ts.map

package/dist/coordination/failure-modes.d.ts.map

@@ -0,0 +1 @@
+{"version":3,"file":"failure-modes.d.ts","sourceRoot":"","sources":["../../src/coordination/failure-modes.ts"],"names":[],"mappings":"AAEA;;;GAGG;AAEH,oBAAY,WAAW;IACrB,WAAW,gBAAmB;IAC9B,OAAO,YAAmB;IAC1B,cAAc,mBAAmB;IACjC,SAAS,cAAmB;IAC5B,SAAS,cAAmB;IAC5B,cAAc,mBAAmB;IACjC,OAAO,YAAmB;CAC3B;AAED,oEAAoE;AACpE,wBAAgB,eAAe,CAAC,MAAM,EAAE,MAAM,GAAG,IAAI,GAAG,WAAW,CAUlE;AAED;;;GAGG;AACH,eAAO,MAAM,cAAc,EAAE,MAAM,CAAC,WAAW,EAAE,MAAM,CAetD,CAAC"}

package/dist/coordination/failure-modes.js

@@ -0,0 +1,49 @@
+// Copyright 2026 Robert Winter / Complete Ideas
+// SPDX-License-Identifier: Apache-2.0
+/**
+ * FailureMode taxonomy and mutation-hint map for the coordination control layer.
+ * Part of AWM 0.8.1 — additive, no breaking changes.
+ */
+export var FailureMode;
+(function (FailureMode) {
+    FailureMode["AGENT_STALE"] = "agent_stale";
+    FailureMode["TIMEOUT"] = "timeout";
+    FailureMode["OUTPUT_INVALID"] = "output_invalid";
+    FailureMode["TEST_FAIL"] = "test_fail";
+    FailureMode["LINT_FAIL"] = "lint_fail";
+    FailureMode["MERGE_CONFLICT"] = "merge_conflict";
+    FailureMode["UNKNOWN"] = "unknown";
+})(FailureMode || (FailureMode = {}));
+/** Classify a failure result string into one of the known modes. */
+export function classifyFailure(result) {
+    if (!result)
+        return FailureMode.UNKNOWN;
+    const r = result.toLowerCase();
+    if (r.includes('stale') || r.includes('disconnected'))
+        return FailureMode.AGENT_STALE;
+    if (r.includes('timeout') || r.includes('timed out'))
+        return FailureMode.TIMEOUT;
+    if (r.includes('json') || r.includes('schema') || r.includes('parse'))
+        return FailureMode.OUTPUT_INVALID;
+    if (r.includes('test fail') || r.includes('vitest') || r.includes('jest'))
+        return FailureMode.TEST_FAIL;
+    if (r.includes('lint') || r.includes('eslint') || r.includes('typecheck'))
+        return FailureMode.LINT_FAIL;
+    if (r.includes('conflict'))
+        return FailureMode.MERGE_CONFLICT;
+    return FailureMode.UNKNOWN;
+}
+/**
+ * Corrective guidance injected into the task description on retry.
+ * Each hint is written in the vocabulary the next worker will read.
+ */
+export const MUTATION_HINTS = {
+    [FailureMode.AGENT_STALE]: 'Previous worker disconnected before completion. Resume from last known state; check git status before re-running destructive commands.',
+    [FailureMode.TIMEOUT]: 'Previous attempt timed out. Break work into smaller commits; report progress every 5 minutes.',
+    [FailureMode.OUTPUT_INVALID]: 'Previous output failed validation. Return a single fenced code block; verify JSON parses before submitting.',
+    [FailureMode.TEST_FAIL]: 'Previous attempt left tests failing. Run vitest before completion; do NOT mark complete if any test fails.',
+    [FailureMode.LINT_FAIL]: 'Previous attempt had lint/typecheck errors. Run pnpm typecheck and pnpm lint before completion.',
+    [FailureMode.MERGE_CONFLICT]: 'Previous attempt left merge conflicts unresolved. git pull --rebase, resolve, then re-attempt.',
+    [FailureMode.UNKNOWN]: 'Previous attempt failed for an unclassified reason. Investigate the prior result before re-running.',
+};
+//# sourceMappingURL=failure-modes.js.map

package/dist/coordination/failure-modes.js.map

@@ -0,0 +1 @@
+{"version":3,"file":"failure-modes.js","sourceRoot":"","sources":["../../src/coordination/failure-modes.ts"],"names":[],"mappings":"AAAA,gDAAgD;AAChD,sCAAsC;AACtC;;;GAGG;AAEH,MAAM,CAAN,IAAY,WAQX;AARD,WAAY,WAAW;IACrB,0CAA8B,CAAA;IAC9B,kCAA0B,CAAA;IAC1B,gDAAiC,CAAA;IACjC,sCAA4B,CAAA;IAC5B,sCAA4B,CAAA;IAC5B,gDAAiC,CAAA;IACjC,kCAA0B,CAAA;AAC5B,CAAC,EARW,WAAW,KAAX,WAAW,QAQtB;AAED,oEAAoE;AACpE,MAAM,UAAU,eAAe,CAAC,MAAqB;IACnD,IAAI,CAAC,MAAM;QAAE,OAAO,WAAW,CAAC,OAAO,CAAC;IACxC,MAAM,CAAC,GAAG,MAAM,CAAC,WAAW,EAAE,CAAC;IAC/B,IAAI,CAAC,CAAC,QAAQ,CAAC,OAAO,CAAC,IAAI,CAAC,CAAC,QAAQ,CAAC,cAAc,CAAC;QAAE,OAAO,WAAW,CAAC,WAAW,CAAC;IACtF,IAAI,CAAC,CAAC,QAAQ,CAAC,SAAS,CAAC,IAAI,CAAC,CAAC,QAAQ,CAAC,WAAW,CAAC;QAAG,OAAO,WAAW,CAAC,OAAO,CAAC;IAClF,IAAI,CAAC,CAAC,QAAQ,CAAC,MAAM,CAAC,IAAI,CAAC,CAAC,QAAQ,CAAC,QAAQ,CAAC,IAAI,CAAC,CAAC,QAAQ,CAAC,OAAO,CAAC;QAAE,OAAO,WAAW,CAAC,cAAc,CAAC;IACzG,IAAI,CAAC,CAAC,QAAQ,CAAC,WAAW,CAAC,IAAI,CAAC,CAAC,QAAQ,CAAC,QAAQ,CAAC,IAAI,CAAC,CAAC,QAAQ,CAAC,MAAM,CAAC;QAAE,OAAO,WAAW,CAAC,SAAS,CAAC;IACxG,IAAI,CAAC,CAAC,QAAQ,CAAC,MAAM,CAAC,IAAI,CAAC,CAAC,QAAQ,CAAC,QAAQ,CAAC,IAAI,CAAC,CAAC,QAAQ,CAAC,WAAW,CAAC;QAAE,OAAO,WAAW,CAAC,SAAS,CAAC;IACxG,IAAI,CAAC,CAAC,QAAQ,CAAC,UAAU,CAAC;QAA8B,OAAO,WAAW,CAAC,cAAc,CAAC;IAC1F,OAAO,WAAW,CAAC,OAAO,CAAC;AAC7B,CAAC;AAED;;;GAGG;AACH,MAAM,CAAC,MAAM,cAAc,GAAgC;IACzD,CAAC,WAAW,CAAC,WAAW,CAAC,EACvB,wIAAwI;IAC1I,CAAC,WAAW,CAAC,OAAO,CAAC,EACnB,+FAA+F;IACjG,CAAC,WAAW,CAAC,cAAc,CAAC,EAC1B,6GAA6G;IAC/G,CAAC,WAAW,CAAC,SAAS,CAAC,EACrB,4GAA4G;IAC9G,CAAC,WAAW,CAAC,SAAS,CAAC,EACrB,iGAAiG;IACnG,CAAC,WAAW,CAAC,cAAc,CAAC,EAC1B,gGAAgG;IAClG,CAAC,WAAW,CAAC,OAAO,CAAC,EACnB,qGAAqG;CACxG,CAAC"}

package/dist/coordination/peer-decisions.js

@@ -44,14 +44,14 @@ export function queryPeerDecisions(db, selfName, query, windowHours = 1, limit =
         const safeHours = Math.max(0.1, Math.min(windowHours, 168)); // 6min – 1 week
         const safeLimit = Math.max(1, Math.min(limit, 50));
         const keywords = extractKeywords(query);
-        const baseWhere = `
-      d.created_at >= datetime('now', '-${safeHours} hours')
-      AND a.name != ?
+        const baseWhere = `
+      d.created_at >= datetime('now', '-${safeHours} hours')
+      AND a.name != ?
     `;
-        const select = `
-      SELECT d.id, a.name AS author_name, d.assignment_id, d.tags, d.summary, d.created_at
-      FROM coord_decisions d
-      JOIN coord_agents a ON d.author_id = a.id
+        const select = `
+      SELECT d.id, a.name AS author_name, d.assignment_id, d.tags, d.summary, d.created_at
+      FROM coord_decisions d
+      JOIN coord_agents a ON d.author_id = a.id
     `;
         if (keywords.length === 0) {
             // No useful keywords — return the N most recent decisions from other agents

package/dist/coordination/routes.d.ts.map

@@ -1 +1 @@
-{"version":3,"file":"routes.d.ts","sourceRoot":"","sources":["../../src/coordination/routes.ts"],"names":[],"mappings":"AAEA;;;;GAIG;AAEH,OAAO,KAAK,EAAE,eAAe,EAAE,MAAM,SAAS,CAAC;AAC/C,OAAO,KAAK,QAAQ,MAAM,gBAAgB,CAAC;AAC3C,OAAO,KAAK,EAAE,WAAW,EAAE,MAAM,sBAAsB,CAAC;AA6ExD,wBAAgB,0BAA0B,CAAC,GAAG,EAAE,eAAe,EAAE,EAAE,EAAE,QAAQ,CAAC,QAAQ,EAAE,KAAK,CAAC,EAAE,WAAW,EAAE,QAAQ,CAAC,EAAE,OAAO,aAAa,EAAE,oBAAoB,GAAG,IAAI,CAu9DxK"}
+{"version":3,"file":"routes.d.ts","sourceRoot":"","sources":["../../src/coordination/routes.ts"],"names":[],"mappings":"AAEA;;;;GAIG;AAEH,OAAO,KAAK,EAAE,eAAe,EAAE,MAAM,SAAS,CAAC;AAC/C,OAAO,KAAK,QAAQ,MAAM,gBAAgB,CAAC;AAC3C,OAAO,KAAK,EAAE,WAAW,EAAE,MAAM,sBAAsB,CAAC;AA+ExD,wBAAgB,0BAA0B,CAAC,GAAG,EAAE,eAAe,EAAE,EAAE,EAAE,QAAQ,CAAC,QAAQ,EAAE,KAAK,CAAC,EAAE,WAAW,EAAE,QAAQ,CAAC,EAAE,OAAO,aAAa,EAAE,oBAAoB,GAAG,IAAI,CA4gExK"}

package/dist/coordination/routes.js

@@ -7,7 +7,8 @@
  */
 import { randomUUID } from 'node:crypto';
 import { checkinSchema, checkoutSchema, pulseSchema, nextSchema, assignCreateSchema, assignmentQuerySchema, assignmentClaimSchema, assignmentUpdateSchema, assignmentIdParamSchema, assignmentsListSchema, reassignSchema, lockAcquireSchema, lockReleaseSchema, commandCreateSchema, commandWaitQuerySchema, findingCreateSchema, findingsQuerySchema, findingIdParamSchema, findingUpdateSchema, decisionsQuerySchema, decisionCreateSchema, eventsQuerySchema, staleQuerySchema, workersQuerySchema, agentIdParamSchema, timelineQuerySchema, channelRegisterSchema, channelDeregisterSchema, channelPushSchema, } from './schemas.js';
-import { detectStale, cleanupStale } from './stale.js';
+import { detectStale, cleanupStale, retryOrFailAssignment } from './stale.js';
+import { recordSuccess, recordFailure as circuitRecordFailure, isAvailable } from './circuit-breaker.js';
 /** Pretty timestamp for coordination logs. */
 function ts() {
     return new Date().toLocaleTimeString('en-GB', { hour12: false });
@@ -315,6 +316,10 @@ export function registerCoordinationRoutes(app, db, store, eventBus) {
             const pending = reserved ?? (agentWorkspace
                 ? db.prepare(`SELECT * FROM coord_assignments WHERE status = 'pending' AND agent_id IS NULL AND (workspace = ? OR workspace IS NULL) ${blockedFilter} ORDER BY priority DESC, created_at ASC LIMIT 1`).get(agentWorkspace)
                 : db.prepare(`SELECT * FROM coord_assignments WHERE status = 'pending' AND agent_id IS NULL ${blockedFilter} ORDER BY priority DESC, created_at ASC LIMIT 1`).get());
+            // Circuit breaker: refuse assignment if worker is in open state
+            if (pending && !isAvailable(db, agentId)) {
+                return reply.code(423).send({ status: 'idle', assignment: null, circuit_open: true, reason: 'circuit_open' });
+            }
             if (pending) {
                 const claimed = db.prepare(`UPDATE coord_assignments SET agent_id = ?, status = 'assigned', started_at = datetime('now') WHERE id = ? AND status = 'pending'`).run(agentId, pending.id);
                 if (claimed.changes > 0) {
@@ -516,6 +521,10 @@ export function registerCoordinationRoutes(app, db, store, eventBus) {
         const pending = reserved ?? (agentWorkspace
             ? db.prepare(`SELECT * FROM coord_assignments WHERE status = 'pending' AND agent_id IS NULL AND (workspace = ? OR workspace IS NULL) ${blockedFilter} ORDER BY priority DESC, created_at ASC LIMIT 1`).get(agentWorkspace)
             : db.prepare(`SELECT * FROM coord_assignments WHERE status = 'pending' AND agent_id IS NULL ${blockedFilter} ORDER BY priority DESC, created_at ASC LIMIT 1`).get());
+        // Circuit breaker: refuse assignment if worker is in open state
+        if (pending && !isAvailable(db, agentId)) {
+            return reply.code(423).send({ assignment: null, circuit_open: true, reason: 'circuit_open' });
+        }
         if (pending) {
             const claimed = db.prepare(`UPDATE coord_assignments SET agent_id = ?, status = 'assigned', started_at = datetime('now') WHERE id = ? AND status = 'pending'`).run(agentId, pending.id);
             if (claimed.changes > 0) {
@@ -598,6 +607,13 @@ export function registerCoordinationRoutes(app, db, store, eventBus) {
         if (['completed', 'failed'].includes(status)) {
             coordLog(`${assignInfo?.agent_name ?? 'unknown'} ${status}: ${assignInfo?.task?.slice(0, 80) ?? id}`);
         }
+        // Circuit breaker: track success/failure per worker
+        if (assignInfo?.agent_id) {
+            if (status === 'completed')
+                recordSuccess(db, assignInfo.agent_id);
+            else if (status === 'failed')
+                circuitRecordFailure(db, assignInfo.agent_id);
+        }
         // Emit events
         eventBus?.emit('assignment.updated', { assignmentId: id, agentId: assignInfo?.agent_id ?? null, status, result });
         if (status === 'completed') {
@@ -722,6 +738,29 @@ export function registerCoordinationRoutes(app, db, store, eventBus) {
             return reply.code(400).send({ error: gate.error });
         return reply.send({ ok: true });
     });
+    /** POST /assignment/:id/fail — Worker voluntarily fails an assignment with retry logic.
+     *  Body: { result: string, mode?: FailureMode }
+     *  Returns: { outcome: 'retried' | 'failed', attempt_count, last_failure_mode }
+     */
+    app.post('/assignment/:id/fail', async (req, reply) => {
+        const { id } = assignmentIdParamSchema.parse(req.params);
+        const body = req.body;
+        const result = body?.result ?? 'worker-initiated failure';
+        const mode = body?.mode;
+        const row = db.prepare(`SELECT id, agent_id, status FROM coord_assignments WHERE id = ?`).get(id);
+        if (!row)
+            return reply.code(404).send({ error: 'assignment not found' });
+        if (['completed', 'failed'].includes(row.status)) {
+            return reply.code(400).send({ error: `cannot fail a ${row.status} assignment` });
+        }
+        const agentId = row.agent_id ?? 'unknown';
+        const outcome = retryOrFailAssignment(db, id, agentId, result, mode);
+        // Circuit breaker: record failure on voluntary fail
+        if (agentId !== 'unknown')
+            circuitRecordFailure(db, agentId);
+        const updated = db.prepare(`SELECT attempt_count, last_failure_mode FROM coord_assignments WHERE id = ?`).get(id);
+        return reply.send({ ok: true, outcome, attempt_count: updated?.attempt_count ?? 0, last_failure_mode: updated?.last_failure_mode ?? null });
+    });
     app.put('/assignment/:id', async (req, reply) => {
         const { id } = assignmentIdParamSchema.parse(req.params);
         const parsed = assignmentUpdateSchema.safeParse(req.body);
@@ -1423,6 +1462,10 @@ export function registerCoordinationRoutes(app, db, store, eventBus) {
         const agent = db.prepare(`SELECT name, workspace FROM coord_agents WHERE id = ?`).get(agentId);
         if (!agent)
             return reply.status(404).send({ error: 'Agent not found' });
+        // Circuit breaker: refuse push to open-circuit workers
+        if (!isAvailable(db, agentId)) {
+            return reply.status(423).send({ error: 'circuit_open', reason: 'Worker circuit is open — too many consecutive failures. Try again after 30s or after a successful assignment.' });
+        }
         // Try live channel delivery first
         const session = db.prepare(`SELECT agent_id, channel_id FROM coord_channel_sessions WHERE agent_id = ? AND status = 'connected'`).get(agentId);
         if (session) {