@cabane/companion 0.6.68 → 0.6.70

package/dist/cli.js

@@ -2284,6 +2284,7 @@ import { randomUUID as randomUUID2 } from "crypto";
 // src/api.ts
 var RETRY_BACKOFF_MS = [250, 750];
 var ACTIVE_RUN_OUTBOX_SEQ = 0;
+var LEASE_CHECK_TIMEOUT_MS = 1e4;
 function activeRunOutboxKey(conversationId, agentId) {
   return `activerun:${conversationId}:${agentId}`;
 }
@@ -2473,6 +2474,52 @@ var CabaneApi = class {
       `/api/workspaces/${workspaceId}/conversations/${conversationId}/participants/agents/${agentId}/turn-intent?${q}`
     );
   }
+  // CT1292: RENEW THE TURN'S LEASE — ask the server whether this turn is still
+  // running. The dispatcher calls this on a cadence for the life of the SDK loop,
+  // and out of cadence the moment a commit is refused, so a loop whose turn was
+  // terminalized underneath it stops instead of running to the six-hour cap.
+  //
+  // Single-shot and NEVER retried, on purpose. A renewal is a question about
+  // right now, not a durable write: a failed one has no answer worth replaying,
+  // and the next tick asks again in a few seconds anyway. Retrying would only
+  // stretch one tick over the next one's slot.
+  //
+  // ALL FAILURE IS `unknown`, and this is the safety property the whole feature
+  // rests on. The only thing that may stop a running loop is the server SAYING
+  // the turn is over; a 500, a dropped connection, an expired token or a body we
+  // cannot parse says nothing about the turn, and reading any of them as death
+  // would kill healthy turns every time the API hiccups — strictly worse than the
+  // gap this closes. So the failure direction is fixed here, at the boundary,
+  // rather than left to each caller to remember.
+  //
+  // AND IT IS DEADLINED (Codo's blocking finding #1). `fetch` has no default
+  // timeout, so a request that neither resolves nor rejects — a black-holed
+  // connection, a half-open socket surviving a network change — would hang this
+  // promise forever. That is not merely a slow probe: the caller single-flights
+  // on it, so one stalled request latches the check permanently, every later
+  // cadence tick returns immediately, and the loop this exists to stop runs to
+  // the six-hour cap even after connectivity comes back — the exact defect the
+  // task is about, reintroduced through the mechanism meant to close it.
+  //
+  // The deadline turns that into an ordinary `unknown`: the request rejects, the
+  // in-flight latch clears, and the NEXT tick asks again. Well under the renewal
+  // cadence so a stalled probe is finished with before its successor is due.
+  async checkTurnLease(workspaceId, conversationId, agentId, turnId) {
+    const q = `turnId=${encodeURIComponent(turnId)}`;
+    try {
+      const res = await this.request(
+        "GET",
+        `/api/workspaces/${workspaceId}/conversations/${conversationId}/participants/agents/${agentId}/turn-lease?${q}`,
+        void 0,
+        {
+          signal: AbortSignal.timeout(this.opts.leaseCheckTimeoutMs ?? LEASE_CHECK_TIMEOUT_MS)
+        }
+      );
+      return res?.state === "running" || res?.state === "ended" ? res.state : "unknown";
+    } catch {
+      return "unknown";
+    }
+  }
   // Flips `active_run_started_at` and optionally captures the SDK session
   // id. The dispatcher hits this twice per turn (now() before the SDK
   // loop; null when it settles) plus once with the session id on the
@@ -3110,7 +3157,9 @@ var turnResultReasonSchema = z6.discriminatedUnion("kind", [
   z6.object({ kind: z6.literal("timeout_idle") }),
   z6.object({ kind: z6.literal("timeout_total") }),
   z6.object({ kind: z6.literal("cancelled") }),
+  z6.object({ kind: z6.literal("lease_lost") }),
   z6.object({ kind: z6.literal("skipped") }),
+  z6.object({ kind: z6.literal("session_start_failed") }),
   z6.object({ kind: z6.literal("workspace_tools_missing") }),
   z6.object({ kind: z6.literal("runtime_error") })
 ]);
@@ -3274,8 +3323,12 @@ function normalizeTurnResultReason(reason) {
       return { kind: "timeout_total" };
     case "cancelled":
       return { kind: "cancelled" };
+    case "lease_lost":
+      return { kind: "lease_lost" };
     case "skipped":
       return { kind: "skipped" };
+    case "session_start_failed":
+      return { kind: "session_start_failed" };
     case "workspace_tools_missing":
       return { kind: "workspace_tools_missing" };
     default:
@@ -3724,8 +3777,9 @@ import {
 
 // packages/agent-runtime/src/claude-code/addendum.ts
 var CLAUDE_CODE_ADDENDUM = [
-  "On this harness the Cabane tools register under MCP prefixes. The workspace tool is",
-  "`mcp__cabane__sdk`; the ancillary tools are `mcp__cabane__list_workspaces`,",
+  "On this harness the Cabane tools register under MCP prefixes. The workspace tools are",
+  "`mcp__cabane__query`, `mcp__cabane__mutate` and `mcp__cabane__describe`; the ancillary",
+  "tools are `mcp__cabane__list_workspaces`,",
   "`mcp__cabane__read_binary`, `mcp__cabane__upload`, `mcp__cabane__begin_upload`,",
   "`mcp__cabane__finalize_upload`, and `mcp__cabane__mint_render_token`; the turn tools are",
   "`mcp__cabane_companion__ask`, `mcp__cabane_companion__reply_to`,",
@@ -4002,7 +4056,7 @@ async function* decodeSdkStream(iter, ctx) {
           };
         }
       } else if (!workspaceProven && (msg.type === "assistant" || msg.type === "user" || msg.type === "result")) {
-        resultReason = "workspace_tools_missing";
+        resultReason = "session_start_failed";
         break;
       } else if (msg.type === "assistant") {
         const assistantErr = msg.error;
@@ -4069,7 +4123,7 @@ async function* decodeSdkStream(iter, ctx) {
   if (ctx.signal.aborted) return;
   if (!workspaceProven) {
     ok = false;
-    resultReason ??= "workspace_tools_missing";
+    resultReason ??= "session_start_failed";
   }
   await flushHeldText(buffer, emit, ok);
   yield* drain(out);
@@ -4096,7 +4150,7 @@ function readMcpInventory(msg) {
 }
 function provesWorkspaceTools(msg) {
   const tools = msg.tools;
-  return Array.isArray(tools) && tools.includes("mcp__cabane__sdk");
+  return Array.isArray(tools) && tools.includes("mcp__cabane__query");
 }
 function isSubscriptionWindow(value) {
   return value === "five_hour" || value === "seven_day" || value === "seven_day_opus" || value === "seven_day_sonnet" || value === "overage";
@@ -4150,6 +4204,47 @@ function buildQueryPrompt(req) {
 }
 
 // packages/agent-runtime/src/claude-code/index.ts
+function attachAbortController(options, signal) {
+  if (options.abortController) return;
+  const sdkAbort = new AbortController();
+  if (signal.aborted) sdkAbort.abort();
+  else signal.addEventListener("abort", () => sdkAbort.abort(), { once: true });
+  options.abortController = sdkAbort;
+}
+function sdkEvents(queryFn, req, signal, built, degraded) {
+  attachAbortController(built.options, signal);
+  const prompt = buildQueryPrompt(req);
+  const iter = queryFn({ prompt, options: built.options });
+  return decodeSdkStream(iter, {
+    signal,
+    cwd: req.local.cwd,
+    resumedSessionId: built.resume,
+    degraded
+  });
+}
+async function* runWithResumeRecovery(queryFn, req, signal, initial, augmentOptions, onWarn, degraded) {
+  let emitted = false;
+  for await (const event of sdkEvents(queryFn, req, signal, initial, degraded)) {
+    const recoverResume = initial.resume !== null && !emitted && event.type === "result" && !event.ok && event.reason === "session_start_failed" && event.mcpInventory?.initReceived === false;
+    if (recoverResume) {
+      if (signal.aborted) return;
+      onWarn?.(
+        `claude-code adapter: stored session ${initial.resume} ended before system/init \u2014 starting one fresh session (CT1293 recovery)`,
+        {
+          cwd: req.local.cwd,
+          reason: "session_start_failed",
+          discardedSessionId: initial.resume
+        }
+      );
+      const freshReq = { ...req, session: null };
+      const fresh = buildClaudeCodeOptions(freshReq, augmentOptions);
+      yield* sdkEvents(queryFn, freshReq, signal, fresh, true);
+      return;
+    }
+    emitted = true;
+    yield event;
+  }
+}
 function createClaudeCodeAdapter(deps = {}) {
   const queryFn = deps.queryFn ?? defaultQuery;
   return {
@@ -4165,20 +4260,15 @@ function createClaudeCodeAdapter(deps = {}) {
         );
       }
       const degraded = freshReason !== void 0 && freshReason !== "no_session";
-      if (!options.abortController) {
-        const sdkAbort = new AbortController();
-        if (signal.aborted) sdkAbort.abort();
-        else signal.addEventListener("abort", () => sdkAbort.abort(), { once: true });
-        options.abortController = sdkAbort;
-      }
-      const prompt = buildQueryPrompt(req);
-      const iter = queryFn({ prompt, options });
-      return decodeSdkStream(iter, {
+      return runWithResumeRecovery(
+        queryFn,
+        req,
         signal,
-        cwd: req.local.cwd,
-        resumedSessionId: resume,
+        { options, resume, freshReason },
+        deps.augmentOptions,
+        deps.onWarn,
         degraded
-      });
+      );
     }
   };
 }
@@ -4222,7 +4312,7 @@ var init = (sessionId, model) => ({
   subtype: "init",
   session_id: sessionId,
   mcp_servers: HEALTHY_MCP_INVENTORY.servers,
-  tools: ["mcp__cabane__sdk"],
+  tools: ["mcp__cabane__query"],
   ...model ? { model } : {}
 });
 var assistantText = (text) => ({
@@ -4700,12 +4790,14 @@ function selectAdapter(registry, runtime) {
 
 // packages/agent-runtime/src/opencode/addendum.ts
 var OPENCODE_ADDENDUM = [
-  "Every tool here goes by its plain name: the workspace tool is `sdk`; the turn tools are",
-  "`ask`, `reply_to`, `wake_me`, `send`, `skip_turn`; the ancillary tools are",
-  "`list_workspaces`, `read_binary`, `upload`, `begin_upload`, `finalize_upload`,",
-  "`mint_render_token`; and the host tools are plain verbs too (`bash`, `read`, `edit`). Mind",
-  "the collision: a bare `read` or `edit` is the HOST tool, acting on the local machine \u2014 a",
-  "workspace read is `cabane.files.read` inside an `sdk` program. If a tool ever appears with",
+  "Every tool here goes by its plain name: the workspace tools are `query`, `mutate` and",
+  "`describe`; the turn tools are `ask`, `reply_to`, `wake_me`, `send`, `skip_turn`; the",
+  "ancillary tools are `list_workspaces`, `read_binary`, `upload`, `begin_upload`,",
+  "`finalize_upload`, `mint_render_token`; and the host tools are plain verbs too (`bash`,",
+  "`read`, `edit`). Mind the collision: a bare `read` or `edit` is the HOST tool, acting on the",
+  "local machine \u2014 a workspace read is `query` with the operation name `files.read`. The",
+  "collision runs the other way too: `read` is a host tool here AND `files.read` is an",
+  "operation name, so keep the bare verb and the dotted name apart. If a tool ever appears with",
   "an `mcp__\u2026__` prefix, that prefix is not part of its name here. One more harness fact: you",
   "can interleave narration with tool calls, but only your final message is recorded as the",
   "turn's reply \u2014 write your closing answer last."
@@ -5674,15 +5766,18 @@ var OPENCODE_CONFORMANCE_FIXTURES = [
 // packages/agent-runtime/src/codex/addendum.ts
 var CODEX_ADDENDUM = [
   "Codex surfaces MCP tools under qualified names, and on some versions defers them. The",
-  "workspace tool is `mcp__cabane__sdk` \u2014 if it isn't in your visible tool list, locate it in",
-  "the deferred-tool inventory and invoke that exact qualified name; never conclude the SDK is",
+  "workspace tools are `mcp__cabane__query`, `mcp__cabane__mutate` and",
+  "`mcp__cabane__describe` \u2014 if they aren't in your visible tool list, locate them in",
+  "the deferred-tool inventory and invoke those exact qualified names; never conclude the",
+  "workspace is",
   "absent without attempting discovery and invocation, and if a call fails, report the recorded",
   "error. The turn tools are `mcp__cabane_companion__ask` / `reply_to` / `wake_me` /",
   "`send` / `skip_turn`, and the ancillary tools",
   "`mcp__cabane__list_workspaces` /",
   "`read_binary` / `upload` / `begin_upload` / `finalize_upload` / `mint_render_token` \u2014 all",
   "possibly deferred too. There is no Cabane `read`/`write`/`search` tool \u2014 those are",
-  "`cabane.*` calls inside your program. One more harness fact: you can interleave narration",
+  "operation NAMES you pass to `query` / `mutate`. One more harness fact: you can interleave",
+  "narration",
   "with tool calls, but only your final message is recorded as the turn's reply \u2014 write your",
   "closing answer last."
 ].join(" ");
@@ -5780,9 +5875,12 @@ function readUsage(ev) {
   const usage = asRecord2(ev.usage);
   if (!usage) return void 0;
   const num = (v) => typeof v === "number" && Number.isFinite(v) ? v : 0;
-  const inputTokens = num(usage.input_tokens) + num(usage.cached_input_tokens);
-  const outputTokens = num(usage.output_tokens) + num(usage.reasoning_output_tokens);
-  return { inputTokens, outputTokens };
+  return {
+    inputTokens: num(usage.input_tokens),
+    outputTokens: num(usage.output_tokens),
+    cacheReadTokens: num(usage.cached_input_tokens),
+    cacheCreationTokens: num(usage.cache_write_input_tokens)
+  };
 }
 function toStatus(v) {
   return v === "completed" || v === "failed" ? v : "in_progress";
@@ -6344,24 +6442,14 @@ var errorItem = (id, message) => ({
   type: "item.completed",
   item: { id, type: "error", message }
 });
-var SDK_SUCCESS = { ok: true, result: { source: "native" }, logs: "", truncated: false };
-var SDK_FAILURE = {
+var OP_SUCCESS = { ok: true, result: { source: "native" } };
+var OP_FAILURE = {
   ok: false,
-  result: null,
-  preview: '{"rows":[1,2,3]}',
-  logs: "",
-  truncated: false,
-  error: { kind: "output_too_large", bytes: 169167, limit: 65536 },
-  warnings: [],
-  committed: {
-    calls: [
-      {
-        method: "conversations.update",
-        args: { id: "conv-1", opts: { title: "Renamed before failure" } }
-      }
-    ]
-  },
-  inFlight: null
+  error: {
+    kind: "output_too_large",
+    code: "output_too_large",
+    message: "files.read returned more than the result ceiling"
+  }
 };
 var sessionEvent3 = (threadId, cwd = DIR2, degraded = false, promptFingerprint = fingerprintPrompt(PROMPT)) => ({
   type: "session",
@@ -6372,6 +6460,11 @@ var CODEX_CONFORMANCE_FIXTURES = [
   {
     // A plain text reply: the held agent message seals as the terminal final at
     // `turn.completed`; the usage counters ride the `result`.
+    //
+    // CT1283: the counters NEST — `cached_input_tokens` is a slice of
+    // `input_tokens`, `reasoning_output_tokens` a slice of `output_tokens` — so
+    // input is 100 (not 120) and output is 50 (not 60), with the cache read broken
+    // out beside them. This fixture pinned the double-count before CT1283.
     name: "clean turn",
     request: makeRequest3(),
     nativeStream: [
@@ -6380,6 +6473,7 @@ var CODEX_CONFORMANCE_FIXTURES = [
       turnCompleted({
         input_tokens: 100,
         cached_input_tokens: 20,
+        cache_write_input_tokens: 5,
         output_tokens: 50,
         reasoning_output_tokens: 10
       })
@@ -6387,7 +6481,16 @@ var CODEX_CONFORMANCE_FIXTURES = [
     expected: [
       sessionEvent3(NEW_THREAD_ID),
       { type: "text", body: "Hello there.", terminal: true },
-      { type: "result", ok: true, usage: { inputTokens: 120, outputTokens: 60 } }
+      {
+        type: "result",
+        ok: true,
+        usage: {
+          inputTokens: 100,
+          outputTokens: 50,
+          cacheReadTokens: 20,
+          cacheCreationTokens: 5
+        }
+      }
     ]
   },
   {
@@ -6443,30 +6546,30 @@ var CODEX_CONFORMANCE_FIXTURES = [
     ]
   },
   {
-    // CT1211: recorded SDK success shape. Codex exposes the text compatibility
+    // CT1211: recorded MCP success shape. Codex exposes the text compatibility
     // copy and the native object together; normalization must keep the object.
     name: "MCP success prefers native structured content",
     request: makeRequest3(),
     nativeStream: [
       threadStarted(NEW_THREAD_ID),
       toolFrame("item.started", {
-        id: "sdk-success",
+        id: "query-success",
         type: "mcp_tool_call",
         server: "cabane",
-        tool: "sdk",
-        arguments: { code: 'return { source: "native" }' },
+        tool: "query",
+        arguments: { op: "files.read", args: { path: "/ABOUT.md" } },
         status: "in_progress"
       }),
       toolFrame("item.completed", {
-        id: "sdk-success",
+        id: "query-success",
         type: "mcp_tool_call",
         server: "cabane",
-        tool: "sdk",
-        arguments: { code: 'return { source: "native" }' },
+        tool: "query",
+        arguments: { op: "files.read", args: { path: "/ABOUT.md" } },
         status: "completed",
         result: {
-          content: [{ type: "text", text: JSON.stringify(SDK_SUCCESS) }],
-          structured_content: SDK_SUCCESS
+          content: [{ type: "text", text: JSON.stringify(OP_SUCCESS) }],
+          structured_content: OP_SUCCESS
         }
       }),
       turnCompleted()
@@ -6475,20 +6578,20 @@ var CODEX_CONFORMANCE_FIXTURES = [
       sessionEvent3(NEW_THREAD_ID),
       {
         type: "tool",
-        id: "sdk-success",
-        name: "sdk",
+        id: "query-success",
+        name: "query",
         phase: "start",
         summary: "",
-        input: { code: 'return { source: "native" }' }
+        input: { op: "files.read", args: { path: "/ABOUT.md" } }
       },
       {
         type: "tool",
-        id: "sdk-success",
-        name: "sdk",
+        id: "query-success",
+        name: "query",
         phase: "done",
         summary: "",
-        input: { code: 'return { source: "native" }' },
-        result: SDK_SUCCESS
+        input: { op: "files.read", args: { path: "/ABOUT.md" } },
+        result: OP_SUCCESS
       },
       { type: "result", ok: true }
     ]
@@ -6496,28 +6599,28 @@ var CODEX_CONFORMANCE_FIXTURES = [
   {
     // CT1211 incident shape: MCP `isError` keeps the failed phase, while the raw
     // camel-case structured payload survives instead of the generic `error` token.
-    name: "failed SDK call preserves structured receipt while remaining error phase",
+    name: "failed MCP call preserves structured receipt while remaining error phase",
     request: makeRequest3(),
     nativeStream: [
       threadStarted(NEW_THREAD_ID),
       toolFrame("item.started", {
-        id: "sdk-failed",
+        id: "query-failed",
         type: "mcp_tool_call",
         server: "cabane",
-        tool: "sdk",
-        arguments: { code: "return largeResult" },
+        tool: "query",
+        arguments: { op: "files.read", args: { path: "/ABOUT.md" } },
         status: "in_progress"
       }),
       toolFrame("item.completed", {
-        id: "sdk-failed",
+        id: "query-failed",
         type: "mcp_tool_call",
         server: "cabane",
-        tool: "sdk",
-        arguments: { code: "return largeResult" },
+        tool: "query",
+        arguments: { op: "files.read", args: { path: "/ABOUT.md" } },
         status: "failed",
         result: {
-          content: [{ type: "text", text: JSON.stringify(SDK_FAILURE) }],
-          structuredContent: SDK_FAILURE,
+          content: [{ type: "text", text: JSON.stringify(OP_FAILURE) }],
+          structuredContent: OP_FAILURE,
           isError: true
         },
         error: { message: "error" }
@@ -6528,20 +6631,20 @@ var CODEX_CONFORMANCE_FIXTURES = [
       sessionEvent3(NEW_THREAD_ID),
       {
         type: "tool",
-        id: "sdk-failed",
-        name: "sdk",
+        id: "query-failed",
+        name: "query",
         phase: "start",
         summary: "",
-        input: { code: "return largeResult" }
+        input: { op: "files.read", args: { path: "/ABOUT.md" } }
       },
       {
         type: "tool",
-        id: "sdk-failed",
-        name: "sdk",
+        id: "query-failed",
+        name: "query",
         phase: "error",
         summary: "",
-        input: { code: "return largeResult" },
-        result: SDK_FAILURE
+        input: { op: "files.read", args: { path: "/ABOUT.md" } },
+        result: OP_FAILURE
       },
       { type: "result", ok: true }
     ]
@@ -7772,6 +7875,7 @@ var TurnCommitter = class {
         { err: err instanceof Error ? err.message : String(err), hook },
         "dispatcher: transcript callback failed"
       );
+      deps.onCommitFailed?.(err);
     };
     const commit = {
       commitMessage: async ({ body, kind, seq }) => {
@@ -7836,6 +7940,7 @@ var TurnCommitter = class {
           { err: err instanceof Error ? err.message : String(err) },
           "dispatcher: progress-promotion commit failed"
         );
+        deps.onCommitFailed?.(err);
       }
     });
     this.emit = sinkEmitter(this.pump.sink, this.onError);
@@ -7995,6 +8100,7 @@ var SILENT_MARKER_BODY = "(no reply)";
 var DEFAULT_PREPARING_ROW_DELAY_MS = 1500;
 var DEFAULT_AGENT_IDLE_TIMEOUT_MS = 10 * 6e4;
 var DEFAULT_AGENT_TOTAL_TIMEOUT_MS = 6 * 60 * 6e4;
+var DEFAULT_LEASE_RENEWAL_MS = 3e4;
 function checkoutState(cwd) {
   if (!cwd) return { ok: false, reason: "no working directory was resolved for this turn" };
   if (!existsSync10(cwd)) return { ok: false, reason: `the working directory is gone (${cwd})` };
@@ -8027,13 +8133,19 @@ var LEASE_REFUSALS = /* @__PURE__ */ new Set([
   "turn_already_ended",
   "turn_belongs_elsewhere"
 ]);
-function leaseRefusal(err) {
+function apiErrorCode(err) {
   if (!(err instanceof ApiError)) return null;
   const body = err.body;
-  const code = typeof body === "object" && body !== null && typeof body.error === "string" ? body.error : null;
+  return typeof body === "object" && body !== null && typeof body.error === "string" ? body.error : null;
+}
+function leaseRefusal(err) {
+  const code = apiErrorCode(err);
   if (code && LEASE_REFUSALS.has(code)) return code;
   return null;
 }
+function isWriteFenceRefusal(err) {
+  return err instanceof ApiError && err.status === 409 && apiErrorCode(err) === "not_running";
+}
 var ERROR_BODY_LOG_CAP = 2e3;
 function describeErrorBody(body) {
   if (body === void 0 || body === null) return void 0;
@@ -8379,6 +8491,7 @@ ${reason}`,
       if (pair2.size === 0) this.aborts.delete(key);
     };
     let timeoutReason = null;
+    let leaseLost = false;
     try {
       await this.opts.api.setActiveRun(workspaceId, payload.conversationId, payload.agentId, {
         activeRunStartedAt: (/* @__PURE__ */ new Date()).toISOString(),
@@ -8526,6 +8639,8 @@ ${reason}`,
     let latestSessionState = request.session;
     let settledDiagnostics = null;
     let silentMarkerEmitted = false;
+    let noteCommitFailed = () => {
+    };
     const committer = new TurnCommitter({
       api: this.opts.api,
       workspaceId,
@@ -8551,7 +8666,9 @@ ${reason}`,
       replyState,
       // The ledger-derived owed reply, for the runtime's own declaration when
       // the agent doesn't call `reply_to` (resolveDeclaredReplyField).
-      owedReplyMessageId: turnContext.owedReplyMessageId ?? null
+      owedReplyMessageId: turnContext.owedReplyMessageId ?? null,
+      // CT1292: a commit that didn't land may mean this turn's lease is gone.
+      onCommitFailed: (err) => noteCommitFailed(err)
     });
     const usesHttpTurnControl = turnRuntime === "codex" || turnRuntime === "opencode";
     let turnControlIntentFetched = false;
@@ -8631,6 +8748,35 @@ ${reason}`,
     const totalTimer = setTimeout(() => fireTimeout("timeout_total"), totalTimeoutMs);
     totalTimer.unref?.();
     armIdle();
+    const leaseRenewalMs = this.opts.leaseRenewalMs ?? DEFAULT_LEASE_RENEWAL_MS;
+    let leaseCheckInFlight = false;
+    const checkLease = async (trigger) => {
+      if (leaseLost || leaseCheckInFlight || abortController.signal.aborted) return;
+      leaseCheckInFlight = true;
+      try {
+        const state = await this.opts.api.checkTurnLease(
+          workspaceId,
+          payload.conversationId,
+          payload.agentId,
+          turnId
+        );
+        if (state !== "ended") return;
+        if (abortController.signal.aborted) return;
+        leaseLost = true;
+        turnLog.warn(
+          { turnId, trigger },
+          "dispatcher: this turn is no longer running server-side \u2014 aborting the loop"
+        );
+        abortController.abort();
+      } finally {
+        leaseCheckInFlight = false;
+      }
+    };
+    const leaseTimer = setInterval(() => void checkLease("cadence"), leaseRenewalMs);
+    leaseTimer.unref?.();
+    noteCommitFailed = (err) => {
+      if (isWriteFenceRefusal(err)) void checkLease("commit_refused");
+    };
     try {
       for await (const event of adapter.runTurn(request, abortController.signal)) {
         transcript2?.write(event);
@@ -8695,7 +8841,7 @@ ${reason}`,
       }
       if (abortController.signal.aborted) {
         okResult = false;
-        resultReason = timeoutReason ?? "cancelled";
+        resultReason = leaseLost ? "lease_lost" : timeoutReason ?? "cancelled";
       }
       const emptyResultReason = !skipState.skipped && classifyEmptyResult({ ok: okResult, contentBearingEvents, usage: turnUsage });
       if (emptyResultReason) {
@@ -8760,7 +8906,12 @@ ${reason}`,
     } finally {
       if (idleTimer) clearTimeout(idleTimer);
       clearTimeout(totalTimer);
-      const userCancelled = abortController.signal.aborted && timeoutReason === null;
+      clearInterval(leaseTimer);
+      if (leaseLost) {
+        resultReason = "lease_lost";
+        okResult = false;
+      }
+      const userCancelled = abortController.signal.aborted && timeoutReason === null && !leaseLost;
       if (timeoutReason !== null) {
         resultReason = timeoutReason;
         okResult = false;
@@ -8802,7 +8953,7 @@ ${reason}`,
       if (turnResolvedConfig && Object.keys(turnResolvedConfig).length > 0) {
         body.resolvedConfig = turnResolvedConfig;
       }
-      if (!okResult && resultReason && resultReason !== "cancelled" && !userCancelled && !skipState.skipped) {
+      if (!okResult && resultReason && resultReason !== "cancelled" && !userCancelled && !leaseLost && !skipState.skipped) {
         body.errorReason = resultReason.slice(0, 200);
       }
       if (okResult) {
@@ -8815,8 +8966,8 @@ ${reason}`,
         body.sessionWriteRejected = true;
         this.sessionWriteNotified.add(key);
       }
-      const outcome = skipState.skipped ? "skipped" : userCancelled ? "cancelled" : okResult ? "success" : "failure";
-      const diagnosticReason = outcome === "skipped" ? { kind: "skipped" } : outcome === "cancelled" ? { kind: "cancelled" } : outcome === "failure" ? normalizeTurnResultReason(resultReason) : null;
+      const outcome = skipState.skipped ? "skipped" : userCancelled || leaseLost ? "cancelled" : okResult ? "success" : "failure";
+      const diagnosticReason = outcome === "skipped" ? { kind: "skipped" } : outcome === "cancelled" ? leaseLost ? { kind: "lease_lost" } : { kind: "cancelled" } : outcome === "failure" ? normalizeTurnResultReason(resultReason) : null;
       settledDiagnostics = {
         outcome,
         resultReason: diagnosticReason,

package/dist/runtime.js

@@ -1704,6 +1704,7 @@ import { randomUUID as randomUUID2 } from "crypto";
 // src/api.ts
 var RETRY_BACKOFF_MS = [250, 750];
 var ACTIVE_RUN_OUTBOX_SEQ = 0;
+var LEASE_CHECK_TIMEOUT_MS = 1e4;
 function activeRunOutboxKey(conversationId, agentId) {
   return `activerun:${conversationId}:${agentId}`;
 }
@@ -1893,6 +1894,52 @@ var CabaneApi = class {
       `/api/workspaces/${workspaceId}/conversations/${conversationId}/participants/agents/${agentId}/turn-intent?${q}`
     );
   }
+  // CT1292: RENEW THE TURN'S LEASE — ask the server whether this turn is still
+  // running. The dispatcher calls this on a cadence for the life of the SDK loop,
+  // and out of cadence the moment a commit is refused, so a loop whose turn was
+  // terminalized underneath it stops instead of running to the six-hour cap.
+  //
+  // Single-shot and NEVER retried, on purpose. A renewal is a question about
+  // right now, not a durable write: a failed one has no answer worth replaying,
+  // and the next tick asks again in a few seconds anyway. Retrying would only
+  // stretch one tick over the next one's slot.
+  //
+  // ALL FAILURE IS `unknown`, and this is the safety property the whole feature
+  // rests on. The only thing that may stop a running loop is the server SAYING
+  // the turn is over; a 500, a dropped connection, an expired token or a body we
+  // cannot parse says nothing about the turn, and reading any of them as death
+  // would kill healthy turns every time the API hiccups — strictly worse than the
+  // gap this closes. So the failure direction is fixed here, at the boundary,
+  // rather than left to each caller to remember.
+  //
+  // AND IT IS DEADLINED (Codo's blocking finding #1). `fetch` has no default
+  // timeout, so a request that neither resolves nor rejects — a black-holed
+  // connection, a half-open socket surviving a network change — would hang this
+  // promise forever. That is not merely a slow probe: the caller single-flights
+  // on it, so one stalled request latches the check permanently, every later
+  // cadence tick returns immediately, and the loop this exists to stop runs to
+  // the six-hour cap even after connectivity comes back — the exact defect the
+  // task is about, reintroduced through the mechanism meant to close it.
+  //
+  // The deadline turns that into an ordinary `unknown`: the request rejects, the
+  // in-flight latch clears, and the NEXT tick asks again. Well under the renewal
+  // cadence so a stalled probe is finished with before its successor is due.
+  async checkTurnLease(workspaceId, conversationId, agentId, turnId) {
+    const q = `turnId=${encodeURIComponent(turnId)}`;
+    try {
+      const res = await this.request(
+        "GET",
+        `/api/workspaces/${workspaceId}/conversations/${conversationId}/participants/agents/${agentId}/turn-lease?${q}`,
+        void 0,
+        {
+          signal: AbortSignal.timeout(this.opts.leaseCheckTimeoutMs ?? LEASE_CHECK_TIMEOUT_MS)
+        }
+      );
+      return res?.state === "running" || res?.state === "ended" ? res.state : "unknown";
+    } catch {
+      return "unknown";
+    }
+  }
   // Flips `active_run_started_at` and optionally captures the SDK session
   // id. The dispatcher hits this twice per turn (now() before the SDK
   // loop; null when it settles) plus once with the session id on the
@@ -2609,7 +2656,9 @@ var turnResultReasonSchema = z6.discriminatedUnion("kind", [
   z6.object({ kind: z6.literal("timeout_idle") }),
   z6.object({ kind: z6.literal("timeout_total") }),
   z6.object({ kind: z6.literal("cancelled") }),
+  z6.object({ kind: z6.literal("lease_lost") }),
   z6.object({ kind: z6.literal("skipped") }),
+  z6.object({ kind: z6.literal("session_start_failed") }),
   z6.object({ kind: z6.literal("workspace_tools_missing") }),
   z6.object({ kind: z6.literal("runtime_error") })
 ]);
@@ -2773,8 +2822,12 @@ function normalizeTurnResultReason(reason) {
       return { kind: "timeout_total" };
     case "cancelled":
       return { kind: "cancelled" };
+    case "lease_lost":
+      return { kind: "lease_lost" };
     case "skipped":
       return { kind: "skipped" };
+    case "session_start_failed":
+      return { kind: "session_start_failed" };
     case "workspace_tools_missing":
       return { kind: "workspace_tools_missing" };
     default:
@@ -3223,8 +3276,9 @@ import {
 
 // packages/agent-runtime/src/claude-code/addendum.ts
 var CLAUDE_CODE_ADDENDUM = [
-  "On this harness the Cabane tools register under MCP prefixes. The workspace tool is",
-  "`mcp__cabane__sdk`; the ancillary tools are `mcp__cabane__list_workspaces`,",
+  "On this harness the Cabane tools register under MCP prefixes. The workspace tools are",
+  "`mcp__cabane__query`, `mcp__cabane__mutate` and `mcp__cabane__describe`; the ancillary",
+  "tools are `mcp__cabane__list_workspaces`,",
   "`mcp__cabane__read_binary`, `mcp__cabane__upload`, `mcp__cabane__begin_upload`,",
   "`mcp__cabane__finalize_upload`, and `mcp__cabane__mint_render_token`; the turn tools are",
   "`mcp__cabane_companion__ask`, `mcp__cabane_companion__reply_to`,",
@@ -3501,7 +3555,7 @@ async function* decodeSdkStream(iter, ctx) {
           };
         }
       } else if (!workspaceProven && (msg.type === "assistant" || msg.type === "user" || msg.type === "result")) {
-        resultReason = "workspace_tools_missing";
+        resultReason = "session_start_failed";
         break;
       } else if (msg.type === "assistant") {
         const assistantErr = msg.error;
@@ -3568,7 +3622,7 @@ async function* decodeSdkStream(iter, ctx) {
   if (ctx.signal.aborted) return;
   if (!workspaceProven) {
     ok = false;
-    resultReason ??= "workspace_tools_missing";
+    resultReason ??= "session_start_failed";
   }
   await flushHeldText(buffer, emit, ok);
   yield* drain(out);
@@ -3595,7 +3649,7 @@ function readMcpInventory(msg) {
 }
 function provesWorkspaceTools(msg) {
   const tools = msg.tools;
-  return Array.isArray(tools) && tools.includes("mcp__cabane__sdk");
+  return Array.isArray(tools) && tools.includes("mcp__cabane__query");
 }
 function isSubscriptionWindow(value) {
   return value === "five_hour" || value === "seven_day" || value === "seven_day_opus" || value === "seven_day_sonnet" || value === "overage";
@@ -3649,6 +3703,47 @@ function buildQueryPrompt(req) {
 }
 
 // packages/agent-runtime/src/claude-code/index.ts
+function attachAbortController(options, signal) {
+  if (options.abortController) return;
+  const sdkAbort = new AbortController();
+  if (signal.aborted) sdkAbort.abort();
+  else signal.addEventListener("abort", () => sdkAbort.abort(), { once: true });
+  options.abortController = sdkAbort;
+}
+function sdkEvents(queryFn, req, signal, built, degraded) {
+  attachAbortController(built.options, signal);
+  const prompt = buildQueryPrompt(req);
+  const iter = queryFn({ prompt, options: built.options });
+  return decodeSdkStream(iter, {
+    signal,
+    cwd: req.local.cwd,
+    resumedSessionId: built.resume,
+    degraded
+  });
+}
+async function* runWithResumeRecovery(queryFn, req, signal, initial, augmentOptions, onWarn, degraded) {
+  let emitted = false;
+  for await (const event of sdkEvents(queryFn, req, signal, initial, degraded)) {
+    const recoverResume = initial.resume !== null && !emitted && event.type === "result" && !event.ok && event.reason === "session_start_failed" && event.mcpInventory?.initReceived === false;
+    if (recoverResume) {
+      if (signal.aborted) return;
+      onWarn?.(
+        `claude-code adapter: stored session ${initial.resume} ended before system/init \u2014 starting one fresh session (CT1293 recovery)`,
+        {
+          cwd: req.local.cwd,
+          reason: "session_start_failed",
+          discardedSessionId: initial.resume
+        }
+      );
+      const freshReq = { ...req, session: null };
+      const fresh = buildClaudeCodeOptions(freshReq, augmentOptions);
+      yield* sdkEvents(queryFn, freshReq, signal, fresh, true);
+      return;
+    }
+    emitted = true;
+    yield event;
+  }
+}
 function createClaudeCodeAdapter(deps = {}) {
   const queryFn = deps.queryFn ?? defaultQuery;
   return {
@@ -3664,20 +3759,15 @@ function createClaudeCodeAdapter(deps = {}) {
         );
       }
       const degraded = freshReason !== void 0 && freshReason !== "no_session";
-      if (!options.abortController) {
-        const sdkAbort = new AbortController();
-        if (signal.aborted) sdkAbort.abort();
-        else signal.addEventListener("abort", () => sdkAbort.abort(), { once: true });
-        options.abortController = sdkAbort;
-      }
-      const prompt = buildQueryPrompt(req);
-      const iter = queryFn({ prompt, options });
-      return decodeSdkStream(iter, {
+      return runWithResumeRecovery(
+        queryFn,
+        req,
         signal,
-        cwd: req.local.cwd,
-        resumedSessionId: resume,
+        { options, resume, freshReason },
+        deps.augmentOptions,
+        deps.onWarn,
         degraded
-      });
+      );
     }
   };
 }
@@ -3721,7 +3811,7 @@ var init = (sessionId, model) => ({
   subtype: "init",
   session_id: sessionId,
   mcp_servers: HEALTHY_MCP_INVENTORY.servers,
-  tools: ["mcp__cabane__sdk"],
+  tools: ["mcp__cabane__query"],
   ...model ? { model } : {}
 });
 var assistantText = (text) => ({
@@ -4199,12 +4289,14 @@ function selectAdapter(registry, runtime) {
 
 // packages/agent-runtime/src/opencode/addendum.ts
 var OPENCODE_ADDENDUM = [
-  "Every tool here goes by its plain name: the workspace tool is `sdk`; the turn tools are",
-  "`ask`, `reply_to`, `wake_me`, `send`, `skip_turn`; the ancillary tools are",
-  "`list_workspaces`, `read_binary`, `upload`, `begin_upload`, `finalize_upload`,",
-  "`mint_render_token`; and the host tools are plain verbs too (`bash`, `read`, `edit`). Mind",
-  "the collision: a bare `read` or `edit` is the HOST tool, acting on the local machine \u2014 a",
-  "workspace read is `cabane.files.read` inside an `sdk` program. If a tool ever appears with",
+  "Every tool here goes by its plain name: the workspace tools are `query`, `mutate` and",
+  "`describe`; the turn tools are `ask`, `reply_to`, `wake_me`, `send`, `skip_turn`; the",
+  "ancillary tools are `list_workspaces`, `read_binary`, `upload`, `begin_upload`,",
+  "`finalize_upload`, `mint_render_token`; and the host tools are plain verbs too (`bash`,",
+  "`read`, `edit`). Mind the collision: a bare `read` or `edit` is the HOST tool, acting on the",
+  "local machine \u2014 a workspace read is `query` with the operation name `files.read`. The",
+  "collision runs the other way too: `read` is a host tool here AND `files.read` is an",
+  "operation name, so keep the bare verb and the dotted name apart. If a tool ever appears with",
   "an `mcp__\u2026__` prefix, that prefix is not part of its name here. One more harness fact: you",
   "can interleave narration with tool calls, but only your final message is recorded as the",
   "turn's reply \u2014 write your closing answer last."
@@ -5173,15 +5265,18 @@ var OPENCODE_CONFORMANCE_FIXTURES = [
 // packages/agent-runtime/src/codex/addendum.ts
 var CODEX_ADDENDUM = [
   "Codex surfaces MCP tools under qualified names, and on some versions defers them. The",
-  "workspace tool is `mcp__cabane__sdk` \u2014 if it isn't in your visible tool list, locate it in",
-  "the deferred-tool inventory and invoke that exact qualified name; never conclude the SDK is",
+  "workspace tools are `mcp__cabane__query`, `mcp__cabane__mutate` and",
+  "`mcp__cabane__describe` \u2014 if they aren't in your visible tool list, locate them in",
+  "the deferred-tool inventory and invoke those exact qualified names; never conclude the",
+  "workspace is",
   "absent without attempting discovery and invocation, and if a call fails, report the recorded",
   "error. The turn tools are `mcp__cabane_companion__ask` / `reply_to` / `wake_me` /",
   "`send` / `skip_turn`, and the ancillary tools",
   "`mcp__cabane__list_workspaces` /",
   "`read_binary` / `upload` / `begin_upload` / `finalize_upload` / `mint_render_token` \u2014 all",
   "possibly deferred too. There is no Cabane `read`/`write`/`search` tool \u2014 those are",
-  "`cabane.*` calls inside your program. One more harness fact: you can interleave narration",
+  "operation NAMES you pass to `query` / `mutate`. One more harness fact: you can interleave",
+  "narration",
   "with tool calls, but only your final message is recorded as the turn's reply \u2014 write your",
   "closing answer last."
 ].join(" ");
@@ -5279,9 +5374,12 @@ function readUsage(ev) {
   const usage = asRecord2(ev.usage);
   if (!usage) return void 0;
   const num = (v) => typeof v === "number" && Number.isFinite(v) ? v : 0;
-  const inputTokens = num(usage.input_tokens) + num(usage.cached_input_tokens);
-  const outputTokens = num(usage.output_tokens) + num(usage.reasoning_output_tokens);
-  return { inputTokens, outputTokens };
+  return {
+    inputTokens: num(usage.input_tokens),
+    outputTokens: num(usage.output_tokens),
+    cacheReadTokens: num(usage.cached_input_tokens),
+    cacheCreationTokens: num(usage.cache_write_input_tokens)
+  };
 }
 function toStatus(v) {
   return v === "completed" || v === "failed" ? v : "in_progress";
@@ -5843,24 +5941,14 @@ var errorItem = (id, message) => ({
   type: "item.completed",
   item: { id, type: "error", message }
 });
-var SDK_SUCCESS = { ok: true, result: { source: "native" }, logs: "", truncated: false };
-var SDK_FAILURE = {
+var OP_SUCCESS = { ok: true, result: { source: "native" } };
+var OP_FAILURE = {
   ok: false,
-  result: null,
-  preview: '{"rows":[1,2,3]}',
-  logs: "",
-  truncated: false,
-  error: { kind: "output_too_large", bytes: 169167, limit: 65536 },
-  warnings: [],
-  committed: {
-    calls: [
-      {
-        method: "conversations.update",
-        args: { id: "conv-1", opts: { title: "Renamed before failure" } }
-      }
-    ]
-  },
-  inFlight: null
+  error: {
+    kind: "output_too_large",
+    code: "output_too_large",
+    message: "files.read returned more than the result ceiling"
+  }
 };
 var sessionEvent3 = (threadId, cwd = DIR2, degraded = false, promptFingerprint = fingerprintPrompt(PROMPT)) => ({
   type: "session",
@@ -5871,6 +5959,11 @@ var CODEX_CONFORMANCE_FIXTURES = [
   {
     // A plain text reply: the held agent message seals as the terminal final at
     // `turn.completed`; the usage counters ride the `result`.
+    //
+    // CT1283: the counters NEST — `cached_input_tokens` is a slice of
+    // `input_tokens`, `reasoning_output_tokens` a slice of `output_tokens` — so
+    // input is 100 (not 120) and output is 50 (not 60), with the cache read broken
+    // out beside them. This fixture pinned the double-count before CT1283.
     name: "clean turn",
     request: makeRequest3(),
     nativeStream: [
@@ -5879,6 +5972,7 @@ var CODEX_CONFORMANCE_FIXTURES = [
       turnCompleted({
         input_tokens: 100,
         cached_input_tokens: 20,
+        cache_write_input_tokens: 5,
         output_tokens: 50,
         reasoning_output_tokens: 10
       })
@@ -5886,7 +5980,16 @@ var CODEX_CONFORMANCE_FIXTURES = [
     expected: [
       sessionEvent3(NEW_THREAD_ID),
       { type: "text", body: "Hello there.", terminal: true },
-      { type: "result", ok: true, usage: { inputTokens: 120, outputTokens: 60 } }
+      {
+        type: "result",
+        ok: true,
+        usage: {
+          inputTokens: 100,
+          outputTokens: 50,
+          cacheReadTokens: 20,
+          cacheCreationTokens: 5
+        }
+      }
     ]
   },
   {
@@ -5942,30 +6045,30 @@ var CODEX_CONFORMANCE_FIXTURES = [
     ]
   },
   {
-    // CT1211: recorded SDK success shape. Codex exposes the text compatibility
+    // CT1211: recorded MCP success shape. Codex exposes the text compatibility
     // copy and the native object together; normalization must keep the object.
     name: "MCP success prefers native structured content",
     request: makeRequest3(),
     nativeStream: [
       threadStarted(NEW_THREAD_ID),
       toolFrame("item.started", {
-        id: "sdk-success",
+        id: "query-success",
         type: "mcp_tool_call",
         server: "cabane",
-        tool: "sdk",
-        arguments: { code: 'return { source: "native" }' },
+        tool: "query",
+        arguments: { op: "files.read", args: { path: "/ABOUT.md" } },
         status: "in_progress"
       }),
       toolFrame("item.completed", {
-        id: "sdk-success",
+        id: "query-success",
         type: "mcp_tool_call",
         server: "cabane",
-        tool: "sdk",
-        arguments: { code: 'return { source: "native" }' },
+        tool: "query",
+        arguments: { op: "files.read", args: { path: "/ABOUT.md" } },
         status: "completed",
         result: {
-          content: [{ type: "text", text: JSON.stringify(SDK_SUCCESS) }],
-          structured_content: SDK_SUCCESS
+          content: [{ type: "text", text: JSON.stringify(OP_SUCCESS) }],
+          structured_content: OP_SUCCESS
         }
       }),
       turnCompleted()
@@ -5974,20 +6077,20 @@ var CODEX_CONFORMANCE_FIXTURES = [
       sessionEvent3(NEW_THREAD_ID),
       {
         type: "tool",
-        id: "sdk-success",
-        name: "sdk",
+        id: "query-success",
+        name: "query",
         phase: "start",
         summary: "",
-        input: { code: 'return { source: "native" }' }
+        input: { op: "files.read", args: { path: "/ABOUT.md" } }
       },
       {
         type: "tool",
-        id: "sdk-success",
-        name: "sdk",
+        id: "query-success",
+        name: "query",
         phase: "done",
         summary: "",
-        input: { code: 'return { source: "native" }' },
-        result: SDK_SUCCESS
+        input: { op: "files.read", args: { path: "/ABOUT.md" } },
+        result: OP_SUCCESS
       },
       { type: "result", ok: true }
     ]
@@ -5995,28 +6098,28 @@ var CODEX_CONFORMANCE_FIXTURES = [
   {
     // CT1211 incident shape: MCP `isError` keeps the failed phase, while the raw
     // camel-case structured payload survives instead of the generic `error` token.
-    name: "failed SDK call preserves structured receipt while remaining error phase",
+    name: "failed MCP call preserves structured receipt while remaining error phase",
     request: makeRequest3(),
     nativeStream: [
       threadStarted(NEW_THREAD_ID),
       toolFrame("item.started", {
-        id: "sdk-failed",
+        id: "query-failed",
         type: "mcp_tool_call",
         server: "cabane",
-        tool: "sdk",
-        arguments: { code: "return largeResult" },
+        tool: "query",
+        arguments: { op: "files.read", args: { path: "/ABOUT.md" } },
         status: "in_progress"
       }),
       toolFrame("item.completed", {
-        id: "sdk-failed",
+        id: "query-failed",
         type: "mcp_tool_call",
         server: "cabane",
-        tool: "sdk",
-        arguments: { code: "return largeResult" },
+        tool: "query",
+        arguments: { op: "files.read", args: { path: "/ABOUT.md" } },
         status: "failed",
         result: {
-          content: [{ type: "text", text: JSON.stringify(SDK_FAILURE) }],
-          structuredContent: SDK_FAILURE,
+          content: [{ type: "text", text: JSON.stringify(OP_FAILURE) }],
+          structuredContent: OP_FAILURE,
           isError: true
         },
         error: { message: "error" }
@@ -6027,20 +6130,20 @@ var CODEX_CONFORMANCE_FIXTURES = [
       sessionEvent3(NEW_THREAD_ID),
       {
         type: "tool",
-        id: "sdk-failed",
-        name: "sdk",
+        id: "query-failed",
+        name: "query",
         phase: "start",
         summary: "",
-        input: { code: "return largeResult" }
+        input: { op: "files.read", args: { path: "/ABOUT.md" } }
       },
       {
         type: "tool",
-        id: "sdk-failed",
-        name: "sdk",
+        id: "query-failed",
+        name: "query",
         phase: "error",
         summary: "",
-        input: { code: "return largeResult" },
-        result: SDK_FAILURE
+        input: { op: "files.read", args: { path: "/ABOUT.md" } },
+        result: OP_FAILURE
       },
       { type: "result", ok: true }
     ]
@@ -7271,6 +7374,7 @@ var TurnCommitter = class {
         { err: err instanceof Error ? err.message : String(err), hook },
         "dispatcher: transcript callback failed"
       );
+      deps.onCommitFailed?.(err);
     };
     const commit = {
       commitMessage: async ({ body, kind, seq }) => {
@@ -7335,6 +7439,7 @@ var TurnCommitter = class {
           { err: err instanceof Error ? err.message : String(err) },
           "dispatcher: progress-promotion commit failed"
         );
+        deps.onCommitFailed?.(err);
       }
     });
     this.emit = sinkEmitter(this.pump.sink, this.onError);
@@ -7494,6 +7599,7 @@ var SILENT_MARKER_BODY = "(no reply)";
 var DEFAULT_PREPARING_ROW_DELAY_MS = 1500;
 var DEFAULT_AGENT_IDLE_TIMEOUT_MS = 10 * 6e4;
 var DEFAULT_AGENT_TOTAL_TIMEOUT_MS = 6 * 60 * 6e4;
+var DEFAULT_LEASE_RENEWAL_MS = 3e4;
 function checkoutState(cwd) {
   if (!cwd) return { ok: false, reason: "no working directory was resolved for this turn" };
   if (!existsSync10(cwd)) return { ok: false, reason: `the working directory is gone (${cwd})` };
@@ -7526,13 +7632,19 @@ var LEASE_REFUSALS = /* @__PURE__ */ new Set([
   "turn_already_ended",
   "turn_belongs_elsewhere"
 ]);
-function leaseRefusal(err) {
+function apiErrorCode(err) {
   if (!(err instanceof ApiError)) return null;
   const body = err.body;
-  const code = typeof body === "object" && body !== null && typeof body.error === "string" ? body.error : null;
+  return typeof body === "object" && body !== null && typeof body.error === "string" ? body.error : null;
+}
+function leaseRefusal(err) {
+  const code = apiErrorCode(err);
   if (code && LEASE_REFUSALS.has(code)) return code;
   return null;
 }
+function isWriteFenceRefusal(err) {
+  return err instanceof ApiError && err.status === 409 && apiErrorCode(err) === "not_running";
+}
 var ERROR_BODY_LOG_CAP = 2e3;
 function describeErrorBody(body) {
   if (body === void 0 || body === null) return void 0;
@@ -7878,6 +7990,7 @@ ${reason}`,
       if (pair.size === 0) this.aborts.delete(key);
     };
     let timeoutReason = null;
+    let leaseLost = false;
     try {
       await this.opts.api.setActiveRun(workspaceId, payload.conversationId, payload.agentId, {
         activeRunStartedAt: (/* @__PURE__ */ new Date()).toISOString(),
@@ -8025,6 +8138,8 @@ ${reason}`,
     let latestSessionState = request.session;
     let settledDiagnostics = null;
     let silentMarkerEmitted = false;
+    let noteCommitFailed = () => {
+    };
     const committer = new TurnCommitter({
       api: this.opts.api,
       workspaceId,
@@ -8050,7 +8165,9 @@ ${reason}`,
       replyState,
       // The ledger-derived owed reply, for the runtime's own declaration when
       // the agent doesn't call `reply_to` (resolveDeclaredReplyField).
-      owedReplyMessageId: turnContext.owedReplyMessageId ?? null
+      owedReplyMessageId: turnContext.owedReplyMessageId ?? null,
+      // CT1292: a commit that didn't land may mean this turn's lease is gone.
+      onCommitFailed: (err) => noteCommitFailed(err)
     });
     const usesHttpTurnControl = turnRuntime === "codex" || turnRuntime === "opencode";
     let turnControlIntentFetched = false;
@@ -8130,6 +8247,35 @@ ${reason}`,
     const totalTimer = setTimeout(() => fireTimeout("timeout_total"), totalTimeoutMs);
     totalTimer.unref?.();
     armIdle();
+    const leaseRenewalMs = this.opts.leaseRenewalMs ?? DEFAULT_LEASE_RENEWAL_MS;
+    let leaseCheckInFlight = false;
+    const checkLease = async (trigger) => {
+      if (leaseLost || leaseCheckInFlight || abortController.signal.aborted) return;
+      leaseCheckInFlight = true;
+      try {
+        const state = await this.opts.api.checkTurnLease(
+          workspaceId,
+          payload.conversationId,
+          payload.agentId,
+          turnId
+        );
+        if (state !== "ended") return;
+        if (abortController.signal.aborted) return;
+        leaseLost = true;
+        turnLog.warn(
+          { turnId, trigger },
+          "dispatcher: this turn is no longer running server-side \u2014 aborting the loop"
+        );
+        abortController.abort();
+      } finally {
+        leaseCheckInFlight = false;
+      }
+    };
+    const leaseTimer = setInterval(() => void checkLease("cadence"), leaseRenewalMs);
+    leaseTimer.unref?.();
+    noteCommitFailed = (err) => {
+      if (isWriteFenceRefusal(err)) void checkLease("commit_refused");
+    };
     try {
       for await (const event of adapter.runTurn(request, abortController.signal)) {
         transcript?.write(event);
@@ -8194,7 +8340,7 @@ ${reason}`,
       }
       if (abortController.signal.aborted) {
         okResult = false;
-        resultReason = timeoutReason ?? "cancelled";
+        resultReason = leaseLost ? "lease_lost" : timeoutReason ?? "cancelled";
       }
       const emptyResultReason = !skipState.skipped && classifyEmptyResult({ ok: okResult, contentBearingEvents, usage: turnUsage });
       if (emptyResultReason) {
@@ -8259,7 +8405,12 @@ ${reason}`,
     } finally {
       if (idleTimer) clearTimeout(idleTimer);
       clearTimeout(totalTimer);
-      const userCancelled = abortController.signal.aborted && timeoutReason === null;
+      clearInterval(leaseTimer);
+      if (leaseLost) {
+        resultReason = "lease_lost";
+        okResult = false;
+      }
+      const userCancelled = abortController.signal.aborted && timeoutReason === null && !leaseLost;
       if (timeoutReason !== null) {
         resultReason = timeoutReason;
         okResult = false;
@@ -8301,7 +8452,7 @@ ${reason}`,
       if (turnResolvedConfig && Object.keys(turnResolvedConfig).length > 0) {
         body.resolvedConfig = turnResolvedConfig;
       }
-      if (!okResult && resultReason && resultReason !== "cancelled" && !userCancelled && !skipState.skipped) {
+      if (!okResult && resultReason && resultReason !== "cancelled" && !userCancelled && !leaseLost && !skipState.skipped) {
         body.errorReason = resultReason.slice(0, 200);
       }
       if (okResult) {
@@ -8314,8 +8465,8 @@ ${reason}`,
         body.sessionWriteRejected = true;
         this.sessionWriteNotified.add(key);
       }
-      const outcome = skipState.skipped ? "skipped" : userCancelled ? "cancelled" : okResult ? "success" : "failure";
-      const diagnosticReason = outcome === "skipped" ? { kind: "skipped" } : outcome === "cancelled" ? { kind: "cancelled" } : outcome === "failure" ? normalizeTurnResultReason(resultReason) : null;
+      const outcome = skipState.skipped ? "skipped" : userCancelled || leaseLost ? "cancelled" : okResult ? "success" : "failure";
+      const diagnosticReason = outcome === "skipped" ? { kind: "skipped" } : outcome === "cancelled" ? leaseLost ? { kind: "lease_lost" } : { kind: "cancelled" } : outcome === "failure" ? normalizeTurnResultReason(resultReason) : null;
       settledDiagnostics = {
         outcome,
         resultReason: diagnosticReason,

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@cabane/companion",
-  "version": "0.6.68",
+  "version": "0.6.70",
   "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",