@cabane/companion 0.6.110 → 0.6.112

package/dist/cli.js

@@ -2901,9 +2901,27 @@ function describeErrorBody(body) {
   if (text.length === 0) return void 0;
   return text.length > ERROR_BODY_LOG_CAP ? `${text.slice(0, ERROR_BODY_LOG_CAP)}\u2026` : text;
 }
+var CAUSE_TEXT_CAP = 200;
+function describeError(err) {
+  const message = err instanceof Error ? err.message : String(err);
+  const cause = describeCause(err);
+  return cause === void 0 ? message : `${message}: ${cause}`;
+}
+function describeCause(err) {
+  if (!(err instanceof Error)) return void 0;
+  const cause = err.cause;
+  if (cause === void 0 || cause === null) return void 0;
+  if (typeof cause === "object") {
+    const code = cause.code;
+    if (typeof code === "string" && code.length > 0) return code;
+  }
+  const text = cause instanceof Error ? cause.message : typeof cause === "string" ? cause : void 0;
+  if (text === void 0 || text.length === 0) return void 0;
+  return text.length > CAUSE_TEXT_CAP ? `${text.slice(0, CAUSE_TEXT_CAP)}\u2026` : text;
+}
 function apiErrorLogFields(err) {
   const fields = {
-    err: err instanceof Error ? err.message : String(err)
+    err: describeError(err)
   };
   if (err instanceof ApiError) {
     fields.status = err.status;
@@ -2922,6 +2940,7 @@ import { randomUUID as randomUUID4 } from "crypto";
 
 // src/api.ts
 var RETRY_BACKOFF_MS = [250, 750];
+var TURN_CONTEXT_BACKOFF_MS = [250, 750, 2e3, 5e3];
 var ACTIVE_RUN_OUTBOX_SEQ = 0;
 var LEASE_CHECK_TIMEOUT_MS = 1e4;
 function activeRunOutboxKey(conversationId, agentId) {
@@ -2972,14 +2991,26 @@ var CabaneApi = class {
     return parsed;
   }
   async request(method, path, body, opts = {}) {
-    const { signal, retry = false } = opts;
-    const maxAttempts = retry ? RETRY_BACKOFF_MS.length + 1 : 1;
+    const { signal, retry = method === "GET", backoffMs = RETRY_BACKOFF_MS } = opts;
+    const maxAttempts = retry ? backoffMs.length + 1 : 1;
     for (let attempt = 1; ; attempt++) {
       try {
         return await this.attempt(method, path, body, signal);
       } catch (err) {
         if (attempt >= maxAttempts || signal?.aborted || !isRetryable(err)) throw err;
-        await sleep2(RETRY_BACKOFF_MS[attempt - 1], signal);
+        const delayMs = backoffMs[attempt - 1];
+        this.opts.log?.info(
+          {
+            method,
+            path: pathForLog(path),
+            attempt,
+            maxAttempts,
+            delayMs,
+            err: describeError(err)
+          },
+          "companion: API call failed transiently; retrying"
+        );
+        await sleep2(delayMs, signal);
         if (signal?.aborted) throw err;
       }
     }
@@ -3117,9 +3148,16 @@ var CabaneApi = class {
   // CT1543: `episode` is the dispatch event's own, echoed back so the server can
   // tell this frame from a later wake on the same message. Omitted when the
   // event didn't carry one, which is the only thing an older event can mean.
-  getTurnContext(conversationId, messageId2, turnId, resumed, fresh, episode) {
+  // CT1657: `signal` is the turn's own abort. It matters here more than on any
+  // other read, because this call now waits up to ~8s across five attempts: a
+  // Stop landing in that window used to leave the whole schedule running, and
+  // the person watched a turn they had already stopped keep trying.
+  getTurnContext(conversationId, messageId2, turnId, resumed, fresh, episode, signal) {
     const q = `conversationId=${encodeURIComponent(conversationId)}&messageId=${encodeURIComponent(messageId2)}` + (turnId ? `&turnId=${encodeURIComponent(turnId)}` : "") + (resumed ? "&resumed=1" : "") + (fresh ? "&fresh=1" : "") + (episode !== void 0 ? `&episode=${encodeURIComponent(String(episode))}` : "");
-    return this.request("GET", `/api/agent/turn-context?${q}`);
+    return this.request("GET", `/api/agent/turn-context?${q}`, void 0, {
+      backoffMs: this.opts.turnContextBackoffMs ?? TURN_CONTEXT_BACKOFF_MS,
+      ...signal ? { signal } : {}
+    });
   }
   // CT1380: the local half of a resumed turn's seq floor (see
   // `Outbox.maxSeqForTurn`). 0 when no outbox is configured — a one-shot CLI or
@@ -3204,6 +3242,12 @@ var CabaneApi = class {
         `/api/workspaces/${workspaceId}/conversations/${conversationId}/participants/agents/${agentId}/turn-lease?${q}`,
         void 0,
         {
+          // CT1657: the one GET that opts OUT of the default retry. The reasoning
+          // is right above — a failed probe has no answer worth replaying and the
+          // next tick asks again in seconds, so retrying inside this one would
+          // stretch a tick over its successor's slot, which is what the deadline
+          // exists to prevent.
+          retry: false,
           signal: AbortSignal.timeout(this.opts.leaseCheckTimeoutMs ?? LEASE_CHECK_TIMEOUT_MS)
         }
       );
@@ -3395,6 +3439,10 @@ function isRetryable(err) {
 function isAbortError(err) {
   return err instanceof Error && err.name === "AbortError";
 }
+function pathForLog(path) {
+  const q = path.indexOf("?");
+  return q === -1 ? path : path.slice(0, q);
+}
 function sleep2(ms, signal) {
   return new Promise((resolve2) => {
     if (signal?.aborted) return resolve2();
@@ -9569,6 +9617,7 @@ var TurnExecution = class {
   // identity-checked at the supervisor — then clear the active-run flag);
   // everything past the lease settles in `execute()`'s one `finally`.
   async run() {
+    this.supervisor.registerAbort(this.turnId, this.abortController);
     try {
       await this.fetchContext();
       this.gateTrigger();
@@ -9590,6 +9639,8 @@ var TurnExecution = class {
         this.supervisor.releaseAbort(this.turnId, this.abortController);
         const msg = err instanceof Error ? err.message : String(err);
         await this.concludeAdmittedRun("setup_failed", `setup_failed: ${msg}`);
+      } else {
+        this.supervisor.releaseAbort(this.turnId, this.abortController);
       }
       throw err;
     }
@@ -9671,9 +9722,14 @@ var TurnExecution = class {
         turnId,
         this.resumed,
         void 0,
-        payload.episode
+        payload.episode,
+        this.abortController.signal
       );
     } catch (err) {
+      if (this.abortController.signal.aborted) {
+        turnLog.debug("dispatcher: stopped while loading the turn context");
+        throw this.concluded("cancelled");
+      }
       const status2 = err instanceof ApiError ? err.status : 0;
       if (status2 === 404) {
         turnLog.debug(
@@ -9688,9 +9744,8 @@ var TurnExecution = class {
         );
         throw this.concluded("turn_context_not_admitted");
       }
-      const reason = err instanceof Error ? err.message : String(err);
-      turnLog.debug({ err: reason }, "dispatcher: failed to fetch turn context");
-      const fetchReason = `fetch_failed: ${reason}`;
+      turnLog.warn(apiErrorLogFields(err), "dispatcher: failed to fetch turn context");
+      const fetchReason = `fetch_failed: ${describeError(err)}`;
       throw this.concluded(fetchReason, fetchReason);
     }
     this.turnContext = turnContext;
@@ -10053,7 +10108,12 @@ var TurnExecution = class {
         payload.messageId,
         turnId,
         false,
-        true
+        true,
+        void 0,
+        // CT1657: the turn's abort. This recompose happens mid-run with the
+        // adapter waiting on it, so a Stop here must end the wait rather than
+        // hold the turn open for the rest of the retry schedule.
+        this.abortController.signal
       );
       turnLog.debug(
         { reason },
@@ -10065,6 +10125,10 @@ var TurnExecution = class {
         content: context.content
       };
     } catch (err) {
+      if (this.abortController.signal.aborted) {
+        turnLog.debug({ reason }, "dispatcher: stopped while recomposing the turn context");
+        return null;
+      }
       turnLog.error(
         { reason, ...apiErrorLogFields(err) },
         "Couldn't load the full conversation; continuing with the recent messages."

package/dist/runtime.js

@@ -2235,9 +2235,27 @@ function describeErrorBody(body) {
   if (text.length === 0) return void 0;
   return text.length > ERROR_BODY_LOG_CAP ? `${text.slice(0, ERROR_BODY_LOG_CAP)}\u2026` : text;
 }
+var CAUSE_TEXT_CAP = 200;
+function describeError(err) {
+  const message = err instanceof Error ? err.message : String(err);
+  const cause = describeCause(err);
+  return cause === void 0 ? message : `${message}: ${cause}`;
+}
+function describeCause(err) {
+  if (!(err instanceof Error)) return void 0;
+  const cause = err.cause;
+  if (cause === void 0 || cause === null) return void 0;
+  if (typeof cause === "object") {
+    const code = cause.code;
+    if (typeof code === "string" && code.length > 0) return code;
+  }
+  const text = cause instanceof Error ? cause.message : typeof cause === "string" ? cause : void 0;
+  if (text === void 0 || text.length === 0) return void 0;
+  return text.length > CAUSE_TEXT_CAP ? `${text.slice(0, CAUSE_TEXT_CAP)}\u2026` : text;
+}
 function apiErrorLogFields(err) {
   const fields = {
-    err: err instanceof Error ? err.message : String(err)
+    err: describeError(err)
   };
   if (err instanceof ApiError) {
     fields.status = err.status;
@@ -2256,6 +2274,7 @@ import { randomUUID as randomUUID4 } from "crypto";
 
 // src/api.ts
 var RETRY_BACKOFF_MS = [250, 750];
+var TURN_CONTEXT_BACKOFF_MS = [250, 750, 2e3, 5e3];
 var ACTIVE_RUN_OUTBOX_SEQ = 0;
 var LEASE_CHECK_TIMEOUT_MS = 1e4;
 function activeRunOutboxKey(conversationId, agentId) {
@@ -2306,14 +2325,26 @@ var CabaneApi = class {
     return parsed;
   }
   async request(method, path, body, opts = {}) {
-    const { signal, retry = false } = opts;
-    const maxAttempts = retry ? RETRY_BACKOFF_MS.length + 1 : 1;
+    const { signal, retry = method === "GET", backoffMs = RETRY_BACKOFF_MS } = opts;
+    const maxAttempts = retry ? backoffMs.length + 1 : 1;
     for (let attempt = 1; ; attempt++) {
       try {
         return await this.attempt(method, path, body, signal);
       } catch (err) {
         if (attempt >= maxAttempts || signal?.aborted || !isRetryable(err)) throw err;
-        await sleep(RETRY_BACKOFF_MS[attempt - 1], signal);
+        const delayMs = backoffMs[attempt - 1];
+        this.opts.log?.info(
+          {
+            method,
+            path: pathForLog(path),
+            attempt,
+            maxAttempts,
+            delayMs,
+            err: describeError(err)
+          },
+          "companion: API call failed transiently; retrying"
+        );
+        await sleep(delayMs, signal);
         if (signal?.aborted) throw err;
       }
     }
@@ -2451,9 +2482,16 @@ var CabaneApi = class {
   // CT1543: `episode` is the dispatch event's own, echoed back so the server can
   // tell this frame from a later wake on the same message. Omitted when the
   // event didn't carry one, which is the only thing an older event can mean.
-  getTurnContext(conversationId, messageId2, turnId, resumed, fresh, episode) {
+  // CT1657: `signal` is the turn's own abort. It matters here more than on any
+  // other read, because this call now waits up to ~8s across five attempts: a
+  // Stop landing in that window used to leave the whole schedule running, and
+  // the person watched a turn they had already stopped keep trying.
+  getTurnContext(conversationId, messageId2, turnId, resumed, fresh, episode, signal) {
     const q = `conversationId=${encodeURIComponent(conversationId)}&messageId=${encodeURIComponent(messageId2)}` + (turnId ? `&turnId=${encodeURIComponent(turnId)}` : "") + (resumed ? "&resumed=1" : "") + (fresh ? "&fresh=1" : "") + (episode !== void 0 ? `&episode=${encodeURIComponent(String(episode))}` : "");
-    return this.request("GET", `/api/agent/turn-context?${q}`);
+    return this.request("GET", `/api/agent/turn-context?${q}`, void 0, {
+      backoffMs: this.opts.turnContextBackoffMs ?? TURN_CONTEXT_BACKOFF_MS,
+      ...signal ? { signal } : {}
+    });
   }
   // CT1380: the local half of a resumed turn's seq floor (see
   // `Outbox.maxSeqForTurn`). 0 when no outbox is configured — a one-shot CLI or
@@ -2538,6 +2576,12 @@ var CabaneApi = class {
         `/api/workspaces/${workspaceId}/conversations/${conversationId}/participants/agents/${agentId}/turn-lease?${q}`,
         void 0,
         {
+          // CT1657: the one GET that opts OUT of the default retry. The reasoning
+          // is right above — a failed probe has no answer worth replaying and the
+          // next tick asks again in seconds, so retrying inside this one would
+          // stretch a tick over its successor's slot, which is what the deadline
+          // exists to prevent.
+          retry: false,
           signal: AbortSignal.timeout(this.opts.leaseCheckTimeoutMs ?? LEASE_CHECK_TIMEOUT_MS)
         }
       );
@@ -2729,6 +2773,10 @@ function isRetryable(err) {
 function isAbortError(err) {
   return err instanceof Error && err.name === "AbortError";
 }
+function pathForLog(path) {
+  const q = path.indexOf("?");
+  return q === -1 ? path : path.slice(0, q);
+}
 function sleep(ms, signal) {
   return new Promise((resolve2) => {
     if (signal?.aborted) return resolve2();
@@ -8992,6 +9040,7 @@ var TurnExecution = class {
   // identity-checked at the supervisor — then clear the active-run flag);
   // everything past the lease settles in `execute()`'s one `finally`.
   async run() {
+    this.supervisor.registerAbort(this.turnId, this.abortController);
     try {
       await this.fetchContext();
       this.gateTrigger();
@@ -9013,6 +9062,8 @@ var TurnExecution = class {
         this.supervisor.releaseAbort(this.turnId, this.abortController);
         const msg = err instanceof Error ? err.message : String(err);
         await this.concludeAdmittedRun("setup_failed", `setup_failed: ${msg}`);
+      } else {
+        this.supervisor.releaseAbort(this.turnId, this.abortController);
       }
       throw err;
     }
@@ -9094,9 +9145,14 @@ var TurnExecution = class {
         turnId,
         this.resumed,
         void 0,
-        payload.episode
+        payload.episode,
+        this.abortController.signal
       );
     } catch (err) {
+      if (this.abortController.signal.aborted) {
+        turnLog.debug("dispatcher: stopped while loading the turn context");
+        throw this.concluded("cancelled");
+      }
       const status = err instanceof ApiError ? err.status : 0;
       if (status === 404) {
         turnLog.debug(
@@ -9111,9 +9167,8 @@ var TurnExecution = class {
         );
         throw this.concluded("turn_context_not_admitted");
       }
-      const reason = err instanceof Error ? err.message : String(err);
-      turnLog.debug({ err: reason }, "dispatcher: failed to fetch turn context");
-      const fetchReason = `fetch_failed: ${reason}`;
+      turnLog.warn(apiErrorLogFields(err), "dispatcher: failed to fetch turn context");
+      const fetchReason = `fetch_failed: ${describeError(err)}`;
       throw this.concluded(fetchReason, fetchReason);
     }
     this.turnContext = turnContext;
@@ -9476,7 +9531,12 @@ var TurnExecution = class {
         payload.messageId,
         turnId,
         false,
-        true
+        true,
+        void 0,
+        // CT1657: the turn's abort. This recompose happens mid-run with the
+        // adapter waiting on it, so a Stop here must end the wait rather than
+        // hold the turn open for the rest of the retry schedule.
+        this.abortController.signal
       );
       turnLog.debug(
         { reason },
@@ -9488,6 +9548,10 @@ var TurnExecution = class {
         content: context.content
       };
     } catch (err) {
+      if (this.abortController.signal.aborted) {
+        turnLog.debug({ reason }, "dispatcher: stopped while recomposing the turn context");
+        return null;
+      }
       turnLog.error(
         { reason, ...apiErrorLogFields(err) },
         "Couldn't load the full conversation; continuing with the recent messages."

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@cabane/companion",
-  "version": "0.6.110",
+  "version": "0.6.112",
   "type": "module",
   "description": "The Cabane Companion (headless): connect a coding agent on your machine to your Cabane workspace as a responder — drive work against your own codebase, files, and MCP servers without putting any of it in Cabane.",
   "license": "UNLICENSED",