Skip to content

The open-source app everyone uses to manage agents at work

fix(recovery): skip successful-run handoff for recovery-action-driven runs (#9010)

Opaqueea3a5ea7d25f
diff --git a/server/src/services/recovery/successful-run-handoff.test.ts b/server/src/services/recovery/successful-run-handoff.test.ts
index 5aa042b041..552ec84c8e 100644
--- a/server/src/services/recovery/successful-run-handoff.test.ts
+++ b/server/src/services/recovery/successful-run-handoff.test.ts
@@ -207,6 +207,36 @@ describe("successful run handoff decision", () => {
expect(instruction).not.toMatch(/[\u0000-\u0008\u000B-\u001F\u007F]/);
});
+ it("does not queue for a run woken by source_scoped_recovery_action", () => {
+ expect(
+ decide({
+ run: {
+ ...run,
+ contextSnapshot: {
+ issueId: "issue-1",
+ wakeReason: "source_scoped_recovery_action",
+ },
+ } as any,
+ }),
+ ).toEqual({
+ kind: "skip",
+ reason: "recovery action run owns its own follow-up path",
+ });
+ // the recoveryActionId marker alone is also enough (payloads carry it even
+ // when wakeReason is rewritten downstream)
+ expect(
+ decide({
+ run: {
+ ...run,
+ contextSnapshot: { issueId: "issue-1", recoveryActionId: "recovery-action-1" },
+ } as any,
+ }),
+ ).toEqual({
+ kind: "skip",
+ reason: "recovery action run owns its own follow-up path",
+ });
+ });
+
it("does not queue when the issue already has a valid disposition", () => {
expect(decide({ issue: { ...issue, status: "done" } as any })).toEqual({
kind: "skip",
diff --git a/server/src/services/recovery/successful-run-handoff.ts b/server/src/services/recovery/successful-run-handoff.ts
index e6a3533586..fa8bc04a1f 100644
--- a/server/src/services/recovery/successful-run-handoff.ts
+++ b/server/src/services/recovery/successful-run-handoff.ts
@@ -304,6 +304,19 @@ function isCorrectiveHandoffRun(run: HeartbeatRunRow) {
readString(context.wakeReason) === FINISH_SUCCESSFUL_RUN_HANDOFF_REASON;
}
+// A run woken by source_scoped_recovery_action must not become the source of another
+// successful-run handoff. The handoff idempotency key includes sourceRunId, so every
+// succeeding recovery run mints a fresh handoff wake: recovery run → handoff wake →
+// corrective run → new recovery action → recovery run → …, an unbounded ping-pong that
+// never reaches the handoff-exhausted escalation. Recovery runs own their own follow-up
+// path; if the disposition is still missing, the stranded-issue escalation (blocked +
+// exhausted notice) is the designed exit, not another handoff.
+function isRecoveryActionDrivenRun(run: HeartbeatRunRow) {
+ const context = readRecord(run.contextSnapshot);
+ return readString(context.wakeReason) === "source_scoped_recovery_action" ||
+ readString(context.recoveryActionId) !== null;
+}
+
function isIssueMonitorMaintenanceRun(run: HeartbeatRunRow) {
const context = readRecord(run.contextSnapshot);
const wakeReason = readString(context.wakeReason);
@@ -423,6 +436,7 @@ export function decideSuccessfulRunHandoff(input: {
if (run.status !== "succeeded") return { kind: "skip", reason: "source run did not succeed" };
if (isCorrectiveHandoffRun(run)) return { kind: "skip", reason: "source run is already a corrective handoff run" };
+ if (isRecoveryActionDrivenRun(run)) return { kind: "skip", reason: "recovery action run owns its own follow-up path" };
if (isIssueMonitorMaintenanceRun(run)) return { kind: "skip", reason: "issue monitor run owns its own recovery path" };
if (isCommentDrivenWake(run)) return { kind: "skip", reason: "comment-driven wake already owns the next action" };
if (run.issueCommentStatus === "retry_queued" || run.issueCommentStatus === "retry_exhausted") {