@agent-native/core 0.92.11 → 0.93.0

package/dist/agent/run-manager.d.ts

@@ -208,7 +208,11 @@ export declare function getActiveRunForThread(threadId: string): ActiveRun | nul
  * dead even before the server-side stale reap has fired. Returns
  * `lastProgressAt` so the client-side stuck-detector can show a
  * user-visible "this chat looks stuck" affordance when a run is alive
- * (heartbeating) but not actually emitting events.
+ * (heartbeating) but not actually emitting events. Returns
+ * `awaitingRedispatch` so the client's background follow loop can tell a
+ * legitimately-deferred `chainServerDrivenContinuation` successor (recovery
+ * in progress server-side) apart from a genuinely dead run — see this
+ * field's own doc comment below.
  */
 export declare function getActiveRunForThreadAsync(threadId: string): Promise<{
     runId: string;
@@ -227,10 +231,48 @@ export declare function getActiveRunForThreadAsync(threadId: string): Promise<{
      * diagnosable from the client WITHOUT the unreadable bg-fn logs.
      */
     diagStage?: string | null;
+    /**
+     * True exactly when this run is a `chainServerDrivenContinuation` deferral
+     * (dispatch_mode === 'background', never claimed) still inside
+     * `UNCLAIMED_BACKGROUND_RUN_REDISPATCH_BOUND_MS` — the same condition this
+     * function already uses below to skip its own `reapUnclaimedBackgroundRun`.
+     * Surfaced on `/runs/active` (agent-chat-plugin.ts) so
+     * `agent-chat-adapter.ts`'s follow loop can tell "silently deferred,
+     * server-side recovery in progress" apart from "dead" and stop counting the
+     * quiet gap against its idle timeout — see the THREE-SITE INVARIANT comment
+     * below and in agent-chat-plugin.ts / production-agent.ts. Always false for
+     * an in-memory run (that isolate IS the live producer) and for any run that
+     * isn't an unclaimed background dispatch.
+     */
+    awaitingRedispatch: boolean;
+    /**
+     * True exactly when this run's `in_flight_since` marker is set — a tool
+     * call or A2A `agent_call` delegation is open and has not yet resolved
+     * (see `setRunInFlightMarker` / `IN_FLIGHT_RUN_STALE_GRACE_MS` in
+     * run-store.ts). This is the SAME signal `reapIfStale` reads to grant its
+     * bounded stale-reap grace — computed here from the identical
+     * `in_flight_since` column via `getRunByThread`, never re-derived, so the
+     * client and the reaper cannot disagree about what "in flight" means.
+     *
+     * Surfaced on `/runs/active` (agent-chat-plugin.ts) as the
+     * server-authoritative alternative to the client-side proxy
+     * `RunStuckBanner` currently infers from unresolved `tool-call` content
+     * parts in the local message list — that proxy can go stale after a
+     * reconnect or reader-mode replay; this cannot, because it is read fresh
+     * from SQL on every poll.
+     */
+    hasInFlightWork: boolean;
 } | null>;
 /** Get a run by ID */
 export declare function getRun(runId: string): ActiveRun | null;
-/** Explicitly abort a run (e.g. Stop button) */
+/** Explicitly abort a run (e.g. Stop button). */
 export declare function abortRun(runId: string, reason?: string): boolean;
+/**
+ * Abort a run and wait until the cross-isolate SQL state and terminal event
+ * are durable. Request handlers that start recovery immediately after aborting
+ * must use this path; otherwise the recovery POST can race the old row while
+ * it is still marked running.
+ */
+export declare function abortRunDurably(runId: string, reason?: string): Promise<boolean>;
 export { tryClaimRunSlot } from "./run-store.js";
 //# sourceMappingURL=run-manager.d.ts.map

package/dist/agent/run-manager.d.ts.map

@@ -1 +1 @@
-{"version":3,"file":"run-manager.d.ts","sourceRoot":"","sources":["../../src/agent/run-manager.ts"],"names":[],"mappings":"AA0BA,OAAO,KAAK,EAAE,cAAc,EAAE,QAAQ,EAAE,SAAS,EAAE,MAAM,YAAY,CAAC;AAEtE,MAAM,WAAW,SAAS;IACxB,KAAK,EAAE,MAAM,CAAC;IACd,QAAQ,EAAE,MAAM,CAAC;IACjB,6EAA6E;IAC7E,MAAM,EAAE,MAAM,CAAC;IACf,MAAM,EAAE,QAAQ,EAAE,CAAC;IACnB,MAAM,EAAE,SAAS,CAAC;IAClB,WAAW,EAAE,GAAG,CAAC,CAAC,KAAK,EAAE,QAAQ,KAAK,IAAI,CAAC,CAAC;IAC5C,KAAK,EAAE,eAAe,CAAC;IACvB,WAAW,CAAC,EAAE,MAAM,CAAC;IACrB,SAAS,EAAE,MAAM,CAAC;CACnB;AAQD;;;;;;;;;;;;;;;GAeG;AACH,eAAO,MAAM,kCAAkC,QAAS,CAAC;AAEzD;;;;;;;;;;;;;;;GAeG;AACH,eAAO,MAAM,8BAA8B,QAAS,CAAC;AAErD;;;;;;;;;;;;;;GAcG;AACH,eAAO,MAAM,kCAAkC,QAAc,CAAC;AAE9D;;;;;GAKG;AACH,eAAO,MAAM,sCAAsC,QACf,CAAC;AAErC;;;;;;;GAOG;AACH,eAAO,MAAM,yCAAyC,QACT,CAAC;AAE9C;;;;;;;;;;;;;;;;;;;;;;;GAuBG;AACH,eAAO,MAAM,+BAA+B,SAAU,CAAC;AAEvD,qEAAqE;AACrE,eAAO,MAAM,kCAAkC,QAAsB,CAAC;AAEtE;;;;;GAKG;AACH,eAAO,MAAM,gCAAgC,QAA0B,CAAC;AAExE;;;;;GAKG;AACH,eAAO,MAAM,gCAAgC,QAAiB,CAAC;AAE/D,wFAAwF;AACxF,eAAO,MAAM,+BAA+B,MAAM,CAAC;AAEnD,iEAAiE;AACjE,eAAO,MAAM,6BAA6B,MAAM,CAAC;AAEjD;;;GAGG;AACH,eAAO,MAAM,gCAAgC,OAAQ,CAAC;AAEtD,8EAA8E;AAC9E,eAAO,MAAM,+BAA+B,MAAM,CAAC;AAEnD,wBAAgB,4BAA4B,CAC1C,GAAG,EAAE,MAAM,EACX,eAAe,EAAE,MAAM,GACtB,MAAM,CAIR;AAmDD,MAAM,WAAW,eAAe;IAC9B;;2CAEuC;IACvC,aAAa,CAAC,EAAE,MAAM,CAAC;IACvB;4DACwD;IACxD,2BAA2B,CAAC,EAAE,OAAO,CAAC;IACtC;;;+EAG2E;IAC3E,MAAM,CAAC,EAAE,MAAM,CAAC;IAChB;;;;;;;;OAQG;IACH,kBAAkB,CAAC,EAAE,OAAO,CAAC;IAC7B;;;;;OAKG;IACH,mBAAmB,CAAC,EAAE,MAAM,CAAC;IAC7B;;;;OAIG;IACH,YAAY,CAAC,EAAE,YAAY,GAAG,uBAAuB,CAAC;CACvD;AAED,MAAM,WAAW,4BAA4B;IAC3C,gBAAgB,CAAC,EAAE,OAAO,CAAC;IAC3B;;;;;;OAMG;IACH,kBAAkB,CAAC,EAAE,OAAO,CAAC;CAC9B;AA0BD,wBAAgB,uBAAuB,CACrC,UAAU,CAAC,EAAE,MAAM,EACnB,OAAO,CAAC,EAAE,4BAA4B,GACrC,MAAM,CAkCR;AAED,wBAAgB,8BAA8B,IAAI,MAAM,CAOvD;AAED,wBAAgB,4BAA4B,IAAI,MAAM,CAOrD;AAiDD;;;;;;GAMG;AACH,wBAAgB,QAAQ,CACtB,KAAK,EAAE,MAAM,EACb,QAAQ,EAAE,MAAM,EAChB,KAAK,EAAE,CACL,IAAI,EAAE,CAAC,KAAK,EAAE,cAAc,KAAK,IAAI,EACrC,MAAM,EAAE,WAAW,KAChB,OAAO,CAAC,IAAI,CAAC,EAClB,UAAU,CAAC,EAAE,CAAC,GAAG,EAAE,SAAS,KAAK,IAAI,GAAG,OAAO,CAAC,IAAI,CAAC,EACrD,OAAO,CAAC,EAAE,eAAe,GACxB,SAAS,CAwoBX;AAED;;;;;;;GAOG;AACH,wBAAgB,cAAc,CAC5B,KAAK,EAAE,MAAM,EACb,OAAO,EAAE,MAAM,GACd,cAAc,CAAC,UAAU,CAAC,GAAG,IAAI,CAOnC;AA6RD,wEAAwE;AACxE,wBAAgB,qBAAqB,CAAC,QAAQ,EAAE,MAAM,GAAG,SAAS,GAAG,IAAI,CAOxE;AAED;;;;;;;;;GASG;AACH,wBAAsB,0BAA0B,CAAC,QAAQ,EAAE,MAAM,GAAG,OAAO,CAAC;IAC1E,KAAK,EAAE,MAAM,CAAC;IACd,QAAQ,EAAE,MAAM,CAAC;IACjB,MAAM,EAAE,MAAM,CAAC;IACf,MAAM,EAAE,MAAM,CAAC;IACf,WAAW,EAAE,MAAM,CAAC;IACpB,cAAc,EAAE,MAAM,GAAG,IAAI,CAAC;IAC9B,+FAA+F;IAC/F,YAAY,CAAC,EAAE,MAAM,GAAG,IAAI,CAAC;IAC7B,0EAA0E;IAC1E,cAAc,CAAC,EAAE,MAAM,GAAG,IAAI,CAAC;IAC/B;;;;OAIG;IACH,SAAS,CAAC,EAAE,MAAM,GAAG,IAAI,CAAC;CAC3B,GAAG,IAAI,CAAC,CAsIR;AAgBD,sBAAsB;AACtB,wBAAgB,MAAM,CAAC,KAAK,EAAE,MAAM,GAAG,SAAS,GAAG,IAAI,CAEtD;AAED,gDAAgD;AAChD,wBAAgB,QAAQ,CAAC,KAAK,EAAE,MAAM,EAAE,MAAM,GAAE,MAAe,GAAG,OAAO,CAQxE;AAGD,OAAO,EAAE,eAAe,EAAE,MAAM,gBAAgB,CAAC"}
+{"version":3,"file":"run-manager.d.ts","sourceRoot":"","sources":["../../src/agent/run-manager.ts"],"names":[],"mappings":"AA+BA,OAAO,KAAK,EAAE,cAAc,EAAE,QAAQ,EAAE,SAAS,EAAE,MAAM,YAAY,CAAC;AAEtE,MAAM,WAAW,SAAS;IACxB,KAAK,EAAE,MAAM,CAAC;IACd,QAAQ,EAAE,MAAM,CAAC;IACjB,6EAA6E;IAC7E,MAAM,EAAE,MAAM,CAAC;IACf,MAAM,EAAE,QAAQ,EAAE,CAAC;IACnB,MAAM,EAAE,SAAS,CAAC;IAClB,WAAW,EAAE,GAAG,CAAC,CAAC,KAAK,EAAE,QAAQ,KAAK,IAAI,CAAC,CAAC;IAC5C,KAAK,EAAE,eAAe,CAAC;IACvB,WAAW,CAAC,EAAE,MAAM,CAAC;IACrB,SAAS,EAAE,MAAM,CAAC;CACnB;AAQD;;;;;;;;;;;;;;;GAeG;AACH,eAAO,MAAM,kCAAkC,QAAS,CAAC;AAEzD;;;;;;;;;;;;;;;GAeG;AACH,eAAO,MAAM,8BAA8B,QAAS,CAAC;AAErD;;;;;;;;;;;;;;GAcG;AACH,eAAO,MAAM,kCAAkC,QAAc,CAAC;AAE9D;;;;;GAKG;AACH,eAAO,MAAM,sCAAsC,QACf,CAAC;AAErC;;;;;;;GAOG;AACH,eAAO,MAAM,yCAAyC,QACT,CAAC;AAE9C;;;;;;;;;;;;;;;;;;;;;;;GAuBG;AACH,eAAO,MAAM,+BAA+B,SAAU,CAAC;AAEvD,qEAAqE;AACrE,eAAO,MAAM,kCAAkC,QAAsB,CAAC;AAEtE;;;;;GAKG;AACH,eAAO,MAAM,gCAAgC,QAA0B,CAAC;AAExE;;;;;GAKG;AACH,eAAO,MAAM,gCAAgC,QAAiB,CAAC;AAE/D,wFAAwF;AACxF,eAAO,MAAM,+BAA+B,MAAM,CAAC;AAEnD,iEAAiE;AACjE,eAAO,MAAM,6BAA6B,MAAM,CAAC;AAEjD;;;GAGG;AACH,eAAO,MAAM,gCAAgC,OAAQ,CAAC;AAEtD,8EAA8E;AAC9E,eAAO,MAAM,+BAA+B,MAAM,CAAC;AAEnD,wBAAgB,4BAA4B,CAC1C,GAAG,EAAE,MAAM,EACX,eAAe,EAAE,MAAM,GACtB,MAAM,CAIR;AAmDD,MAAM,WAAW,eAAe;IAC9B;;2CAEuC;IACvC,aAAa,CAAC,EAAE,MAAM,CAAC;IACvB;4DACwD;IACxD,2BAA2B,CAAC,EAAE,OAAO,CAAC;IACtC;;;+EAG2E;IAC3E,MAAM,CAAC,EAAE,MAAM,CAAC;IAChB;;;;;;;;OAQG;IACH,kBAAkB,CAAC,EAAE,OAAO,CAAC;IAC7B;;;;;OAKG;IACH,mBAAmB,CAAC,EAAE,MAAM,CAAC;IAC7B;;;;OAIG;IACH,YAAY,CAAC,EAAE,YAAY,GAAG,uBAAuB,CAAC;CACvD;AAED,MAAM,WAAW,4BAA4B;IAC3C,gBAAgB,CAAC,EAAE,OAAO,CAAC;IAC3B;;;;;;OAMG;IACH,kBAAkB,CAAC,EAAE,OAAO,CAAC;CAC9B;AA0BD,wBAAgB,uBAAuB,CACrC,UAAU,CAAC,EAAE,MAAM,EACnB,OAAO,CAAC,EAAE,4BAA4B,GACrC,MAAM,CAkCR;AAED,wBAAgB,8BAA8B,IAAI,MAAM,CAOvD;AAED,wBAAgB,4BAA4B,IAAI,MAAM,CAOrD;AAqDD;;;;;;GAMG;AACH,wBAAgB,QAAQ,CACtB,KAAK,EAAE,MAAM,EACb,QAAQ,EAAE,MAAM,EAChB,KAAK,EAAE,CACL,IAAI,EAAE,CAAC,KAAK,EAAE,cAAc,KAAK,IAAI,EACrC,MAAM,EAAE,WAAW,KAChB,OAAO,CAAC,IAAI,CAAC,EAClB,UAAU,CAAC,EAAE,CAAC,GAAG,EAAE,SAAS,KAAK,IAAI,GAAG,OAAO,CAAC,IAAI,CAAC,EACrD,OAAO,CAAC,EAAE,eAAe,GACxB,SAAS,CAirBX;AAED;;;;;;;GAOG;AACH,wBAAgB,cAAc,CAC5B,KAAK,EAAE,MAAM,EACb,OAAO,EAAE,MAAM,GACd,cAAc,CAAC,UAAU,CAAC,GAAG,IAAI,CAOnC;AA6RD,wEAAwE;AACxE,wBAAgB,qBAAqB,CAAC,QAAQ,EAAE,MAAM,GAAG,SAAS,GAAG,IAAI,CAOxE;AAED;;;;;;;;;;;;;GAaG;AACH,wBAAsB,0BAA0B,CAAC,QAAQ,EAAE,MAAM,GAAG,OAAO,CAAC;IAC1E,KAAK,EAAE,MAAM,CAAC;IACd,QAAQ,EAAE,MAAM,CAAC;IACjB,MAAM,EAAE,MAAM,CAAC;IACf,MAAM,EAAE,MAAM,CAAC;IACf,WAAW,EAAE,MAAM,CAAC;IACpB,cAAc,EAAE,MAAM,GAAG,IAAI,CAAC;IAC9B,+FAA+F;IAC/F,YAAY,CAAC,EAAE,MAAM,GAAG,IAAI,CAAC;IAC7B,0EAA0E;IAC1E,cAAc,CAAC,EAAE,MAAM,GAAG,IAAI,CAAC;IAC/B;;;;OAIG;IACH,SAAS,CAAC,EAAE,MAAM,GAAG,IAAI,CAAC;IAC1B;;;;;;;;;;;;OAYG;IACH,kBAAkB,EAAE,OAAO,CAAC;IAC5B;;;;;;;;;;;;;;;OAeG;IACH,eAAe,EAAE,OAAO,CAAC;CAC1B,GAAG,IAAI,CAAC,CAiKR;AAgBD,sBAAsB;AACtB,wBAAgB,MAAM,CAAC,KAAK,EAAE,MAAM,GAAG,SAAS,GAAG,IAAI,CAEtD;AAUD,iDAAiD;AACjD,wBAAgB,QAAQ,CAAC,KAAK,EAAE,MAAM,EAAE,MAAM,GAAE,MAAe,GAAG,OAAO,CAKxE;AAED;;;;;GAKG;AACH,wBAAsB,eAAe,CACnC,KAAK,EAAE,MAAM,EACb,MAAM,GAAE,MAAe,GACtB,OAAO,CAAC,OAAO,CAAC,CAuBlB;AAGD,OAAO,EAAE,eAAe,EAAE,MAAM,gBAAgB,CAAC"}

package/dist/agent/run-manager.js

@@ -1,7 +1,7 @@
 import { captureError } from "../server/capture-error.js";
-import { isLlmCredentialError } from "./engine/credential-errors.js";
+import { isLlmCredentialError, LLM_MISSING_CREDENTIALS_ERROR_CODE, LLM_MISSING_CREDENTIALS_MESSAGE, } from "./engine/credential-errors.js";
 import { EngineError } from "./engine/types.js";
-import { insertRun, insertRunEvent, updateRunStatusIfRunning, markRunAborted, getRunAbortState, getRunStatus, getRunEventsSince, getRunById, getRunByThread, cleanupOldRuns, updateRunHeartbeat, bumpRunProgress, reapIfStale, reapUnclaimedBackgroundRun, shouldRedispatchUnclaimedBackgroundRun, reconcileTerminalRunFromEvents, ensureTerminalRunEvent, getLastTerminalRunEvent, resolveErroredRunTerminalEvent, setRunError, setRunTerminalReason, } from "./run-store.js";
+import { insertRun, insertRunEvent, updateRunStatusIfRunning, markRunAborted, getRunAbortState, getRunStatus, getRunEventsSince, getRunById, getRunByThread, cleanupOldRuns, updateRunHeartbeat, bumpRunProgress, setRunInFlightMarker, reapIfStale, reapUnclaimedBackgroundRun, shouldRedispatchUnclaimedBackgroundRun, reconcileTerminalRunFromEvents, ensureTerminalRunEvent, getLastTerminalRunEvent, resolveErroredRunTerminalEvent, setRunError, setRunTerminalReason, } from "./run-store.js";
 const activeRuns = new Map();
 const threadToRun = new Map();
 /** How long to keep completed runs in memory before cleanup (5 min) */
@@ -254,6 +254,9 @@ function isTerminalRunEvent(event) {
         event.type === "loop_limit" ||
         event.type === "auto_continue");
 }
+function terminalEventForcesErroredStatus(event) {
+    return event?.type === "error" || event?.type === "missing_api_key";
+}
 function terminalReasonForRun(finalStatus, terminalEvent, abortReason, completionError) {
     if (terminalEvent?.type === "auto_continue") {
         return terminalEvent.reason || "auto_continue";
@@ -452,6 +455,7 @@ export function startRun(runId, threadId, runFn, onComplete, options) {
     let lastRealProgressAt = Date.now();
     let inFlightWorkCount = 0;
     const trackInFlightWork = (event) => {
+        const wasIdle = inFlightWorkCount === 0;
         if (event.type === "tool_start") {
             inFlightWorkCount += 1;
         }
@@ -466,6 +470,25 @@ export function startRun(runId, threadId, runFn, onComplete, options) {
                 inFlightWorkCount = Math.max(0, inFlightWorkCount - 1);
             }
         }
+        else {
+            return; // Not a work-tracking event — no transition possible.
+        }
+        // Mirror the 0<->N transition into SQL so a stale reaper running in a
+        // DIFFERENT isolate (a client's SQL-subscription poll, a sibling
+        // isolate's opportunistic cleanup, a fresh boot's startup sweep) can
+        // grant this demonstrably-alive run a bounded grace even when THIS
+        // isolate's own heartbeat write is failing (e.g. Neon pooler saturation)
+        // — `inFlightWorkCount` itself is in-memory and invisible to those other
+        // isolates. Fire-and-forget: never block event emission on this write,
+        // and a write failure here is no worse than today's behavior (the row
+        // just gets no grace). See `setRunInFlightMarker` / `IN_FLIGHT_RUN_STALE_GRACE_MS`
+        // in run-store.ts for the full reasoning and the bounded-grace derivation.
+        if (wasIdle && inFlightWorkCount > 0) {
+            setRunInFlightMarker(runId, true).catch(() => { });
+        }
+        else if (!wasIdle && inFlightWorkCount === 0) {
+            setRunInFlightMarker(runId, false).catch(() => { });
+        }
     };
     const checkNoProgressBackstop = () => {
         if (noProgressTimeoutMs <= 0)
@@ -717,11 +740,22 @@ export function startRun(runId, threadId, runFn, onComplete, options) {
         //    /runs/active check while we wait for SQL writes to land.
         let completionError = null;
         let terminalPersistenceError = null;
+        const terminalEvent = pendingTerminalEvent?.event ?? null;
         if (onComplete &&
             !(run.status === "aborted" && run.abortReason === "no_progress")) {
             try {
-                const completionRun = pendingTerminalEvent
-                    ? { ...run, events: [...run.events, pendingTerminalEvent] }
+                const completionStatus = run.status !== "aborted" &&
+                    terminalEventForcesErroredStatus(terminalEvent)
+                    ? "errored"
+                    : run.status;
+                const completionRun = pendingTerminalEvent || completionStatus !== run.status
+                    ? {
+                        ...run,
+                        status: completionStatus,
+                        events: pendingTerminalEvent
+                            ? [...run.events, pendingTerminalEvent]
+                            : run.events,
+                    }
                     : run;
                 await onComplete(completionRun);
             }
@@ -736,10 +770,12 @@ export function startRun(runId, threadId, runFn, onComplete, options) {
         //    thread_data.
         const finalStatus = run.status === "aborted"
             ? "aborted"
-            : run.status === "errored" || completionError
+            : run.status === "errored" ||
+                completionError ||
+                terminalEventForcesErroredStatus(terminalEvent)
                 ? "errored"
                 : "completed";
-        const terminalReason = terminalReasonForRun(finalStatus, pendingTerminalEvent?.event ?? null, run.abortReason, completionError);
+        const terminalReason = terminalReasonForRun(finalStatus, terminalEvent, run.abortReason, completionError);
         // 3. Emit the terminal event only after thread_data is durable. Live
         //    SSE clients close on this event and usually fetch thread_data
         //    immediately, so emitting it earlier recreates the final-message
@@ -758,7 +794,8 @@ export function startRun(runId, threadId, runFn, onComplete, options) {
             // re-stamp the seq at emit time (max-seq+1) just below.
             const terminalEvent = finalStatus === "completed"
                 ? (pendingTerminalEvent?.event ?? { type: "done" })
-                : pendingTerminalEvent?.event.type === "error"
+                : pendingTerminalEvent?.event.type === "error" ||
+                    pendingTerminalEvent?.event.type === "missing_api_key"
                     ? pendingTerminalEvent.event
                     : pendingTerminalEvent?.event.type === "auto_continue"
                         ? // The run was checkpointed at a soft-timeout/loop boundary and
@@ -834,9 +871,17 @@ export function startRun(runId, threadId, runFn, onComplete, options) {
         if (finalStatus === "errored") {
             let errorCode;
             let errorDetail;
-            for (let i = run.events.length - 1; i >= 0; i--) {
-                const ev = run.events[i].event;
-                if (ev.type === "error") {
+            const diagnosticEvents = pendingTerminalEvent
+                ? [...run.events, pendingTerminalEvent]
+                : run.events;
+            for (let i = diagnosticEvents.length - 1; i >= 0; i--) {
+                const ev = diagnosticEvents[i].event;
+                if (ev.type === "missing_api_key") {
+                    errorCode = LLM_MISSING_CREDENTIALS_ERROR_CODE;
+                    errorDetail = LLM_MISSING_CREDENTIALS_MESSAGE;
+                    break;
+                }
+                else if (ev.type === "error") {
                     errorCode = ev.errorCode;
                     errorDetail = ev.error ?? ev.details;
                     break;
@@ -1163,7 +1208,11 @@ export function getActiveRunForThread(threadId) {
  * dead even before the server-side stale reap has fired. Returns
  * `lastProgressAt` so the client-side stuck-detector can show a
  * user-visible "this chat looks stuck" affordance when a run is alive
- * (heartbeating) but not actually emitting events.
+ * (heartbeating) but not actually emitting events. Returns
+ * `awaitingRedispatch` so the client's background follow loop can tell a
+ * legitimately-deferred `chainServerDrivenContinuation` successor (recovery
+ * in progress server-side) apart from a genuinely dead run — see this
+ * field's own doc comment below.
  */
 export async function getActiveRunForThreadAsync(threadId) {
     // Check memory first — return both running AND recently-completed runs
@@ -1196,6 +1245,18 @@ export async function getActiveRunForThreadAsync(threadId) {
             dispatchMode: sqlSnapshot?.dispatchMode ?? null,
             terminalReason: sqlSnapshot?.terminalReason ?? null,
             diagStage: sqlSnapshot?.diagStage ?? null,
+            // In-memory means this isolate is the live producer — never the
+            // "deferred, nobody producing" state this flag identifies.
+            awaitingRedispatch: false,
+            // Read from the SAME SQL snapshot the other fields above already read
+            // (`fetchRunThreadSnapshot` -> `getRunByThread`) rather than the
+            // producer's own in-memory `inFlightWorkCount` — this isolate IS the
+            // live producer, but there is no separate in-memory channel wired for
+            // that counter today, and the SQL marker is written on every 0<->N
+            // transition (see run-manager's `trackInFlightWork`), so it is at most
+            // one event behind here — the same tolerance `lastProgressAt` above
+            // already accepts.
+            hasInFlightWork: sqlSnapshot?.inFlightSince != null,
         };
     }
     // Fall back to SQL — also surface recently terminated runs so the client
@@ -1239,10 +1300,14 @@ export async function getActiveRunForThreadAsync(threadId) {
             // follow loop AUTO-continues — and once the redispatch bound is exceeded
             // this reap fires loudly as before. So recovery stays automatic in the
             // common case and loud failure is only moved later, never removed.
-            if (sqlRun.dispatchMode === "background" &&
-                !shouldRedispatchUnclaimedBackgroundRun({
-                    startedAt: sqlRun.startedAt,
-                })) {
+            //
+            // `isUnclaimedBackgroundDispatch` also becomes the `awaitingRedispatch`
+            // wire field below once the still-inside-the-bound check passes — see
+            // this function's doc comment and the THREE-SITE INVARIANT comment in
+            // agent-chat-plugin.ts / production-agent.ts.
+            const isUnclaimedBackgroundDispatch = sqlRun.dispatchMode === "background";
+            const stillInsideRedispatchBound = shouldRedispatchUnclaimedBackgroundRun({ startedAt: sqlRun.startedAt });
+            if (isUnclaimedBackgroundDispatch && !stillInsideRedispatchBound) {
                 const recovered = await reapUnclaimedBackgroundRun(sqlRun.id).catch(() => false);
                 if (recovered)
                     return null;
@@ -1263,6 +1328,11 @@ export async function getActiveRunForThreadAsync(threadId) {
                 dispatchMode: sqlRun.dispatchMode,
                 terminalReason: sqlRun.terminalReason,
                 diagStage: sqlRun.diagStage,
+                awaitingRedispatch: isUnclaimedBackgroundDispatch && stillInsideRedispatchBound,
+                // Same `in_flight_since` column `reapIfStale` (just called above,
+                // and it did NOT reap this row) reads for its own grace decision —
+                // one source of truth, not a second re-derived notion of "in flight".
+                hasInFlightWork: sqlRun.inFlightSince != null,
             };
         }
         if (sqlRun.status === "completed" || sqlRun.status === "errored") {
@@ -1291,6 +1361,10 @@ export async function getActiveRunForThreadAsync(threadId) {
                 dispatchMode: sqlRun.dispatchMode,
                 terminalReason: sqlRun.terminalReason,
                 diagStage: sqlRun.diagStage,
+                // Terminal already — never the "still deferred, running" state.
+                awaitingRedispatch: false,
+                // Terminal already — no live work can still be in flight.
+                hasInFlightWork: false,
             };
         }
     }
@@ -1318,15 +1392,47 @@ async function fetchRunThreadSnapshot(runId, threadId) {
 export function getRun(runId) {
     return activeRuns.get(runId) ?? null;
 }
-/** Explicitly abort a run (e.g. Stop button) */
-export function abortRun(runId, reason = "user") {
+function abortRunInMemory(runId, reason) {
     const run = activeRuns.get(runId);
     if (run) {
         abortInMemoryRun(run, reason);
     }
+    return !!run;
+}
+/** Explicitly abort a run (e.g. Stop button). */
+export function abortRun(runId, reason = "user") {
+    const abortedInMemory = abortRunInMemory(runId, reason);
     // Also mark as aborted in SQL (for cross-isolate abort on Workers)
     markRunAborted(runId, reason).catch(() => { });
-    return !!run;
+    return abortedInMemory;
+}
+/**
+ * Abort a run and wait until the cross-isolate SQL state and terminal event
+ * are durable. Request handlers that start recovery immediately after aborting
+ * must use this path; otherwise the recovery POST can race the old row while
+ * it is still marked running.
+ */
+export async function abortRunDurably(runId, reason = "user") {
+    const abortedInMemory = abortRunInMemory(runId, reason);
+    try {
+        await markRunAborted(runId, reason);
+    }
+    catch (error) {
+        // The local run is already stopped. A transient durable cleanup failure
+        // must not turn the user's Stop/Retry request into a 500 after that
+        // irreversible in-memory outcome. Capture it for repair/reaping and let
+        // the request report the abort it did complete.
+        captureError(error, {
+            route: "/_agent-native/agent-chat/runs/:id/abort",
+            tags: {
+                source: "agent-run-manager",
+                phase: "abort-run",
+            },
+            extra: { runId, reason, abortedInMemory },
+        });
+        console.error("[run-manager] durable abort persistence failed:", error instanceof Error ? error.message : error);
+    }
+    return abortedInMemory;
 }
 // Re-export so callers can avoid importing from run-store directly.
 export { tryClaimRunSlot } from "./run-store.js";