PR #65988: fix: report error when cron subagents produce no final output

• Repository: openclaw/openclaw
• Author: lml2468
• State: open | merged: False
• Link: https://github.com/openclaw/openclaw/pull/65988

Description (problem / solution / changelog)

Summary

Fixes #65950 — Isolated cron sessions falsely report status: ok despite incomplete work.

Root cause: In finalizeTextDelivery, when descendant subagents have been observed but die without producing output (e.g., killed by infrastructure timeout), the code falls into a branch that returns status: "ok" with deliveryAttempted: true. The branch's intended semantics were "subagents are still running, don't deliver yet" — but at that point activeSubagentRuns === 0, meaning the subagents already finished or died. The distinction between "still running" and "died without output" was missing.

Note: isAborted() is NOT the right signal here — OpenClaw's own 60-minute timeout hasn't triggered. The subagents were killed by the infrastructure layer (Docker/VPS wall-clock limit), which does not set the abort controller.

Changes

• src/cron/isolated-agent/delivery-dispatch.ts — The hadDescendants + interim text + zero active subagents branch now returns status: "error" instead of status: "ok". The activeSubagentRuns > 0 branch above it is unchanged (still-running = ok is correct).
• src/cron/isolated-agent/run.ts — After the continuation prompt, if output is still interim, sets runResult.meta.error so the error propagates through resolveCronPayloadOutcome / hasFatalErrorPayload naturally.

Tests

• src/cron/isolated-agent/delivery-dispatch.subagent-died.test.ts (new) — 6 tests: dead subagents → error; still-running subagents → ok; subagents with output → delivers; fallback reply → delivers; multiple dead subagents → error
• src/cron/isolated-agent/delivery-dispatch.double-announce.test.ts (updated) — existing stale-interim test now asserts status: "error" and the error message

All 516 cron tests pass.

Note on infra alignment

This fix handles the observable symptom correctly. As a separate operational recommendation: Docker/VPS wall-clock timeouts should be set to timeoutSeconds + buffer to give OpenClaw's own timeout mechanism a chance to exit cleanly before the infra kills the process.

AI Disclosure

• AI-assisted (Claude Code via OpenClaw)
• Fully tested (516 cron tests pass, build succeeds)
• Root cause verified by tracing complete call chain through source
• Understands what the code does — semantic fix to status judgment, not an abort-path patch

Changed files

• src/cron/isolated-agent/delivery-dispatch.double-announce.test.ts (modified, +11/-2)
• src/cron/isolated-agent/delivery-dispatch.subagent-died.test.ts (added, +287/-0)
• src/cron/isolated-agent/delivery-dispatch.ts (modified, +6/-4)
• src/cron/isolated-agent/run-executor.ts (modified, +29/-0)