PR #78575: fix(tasks): mark restart-interrupted tasks lost

• Repository: openclaw/openclaw
• Author: tmimmanuel
• State: open | merged: False
• Link: https://github.com/openclaw/openclaw/pull/78575

Description (problem / solution / changelog)

Summary

• mark active task records as lost when a forced or timed-out gateway restart proceeds before task drain completes
• keep queued tasks untouched while clearing stale running audit leftovers from interrupted work
• cover the registry helper and restart run-loop paths with focused regressions

Fixes #78463.

Verification

• pnpm exec oxfmt --check --threads=1 src/tasks/task-registry.maintenance.ts src/tasks/task-registry.maintenance.issue-60299.test.ts src/cli/gateway-cli/lifecycle.runtime.ts src/cli/gateway-cli/run-loop.ts src/cli/gateway-cli/run-loop.test.ts
• git diff --check
• pnpm test src/tasks/task-registry.maintenance.issue-60299.test.ts src/cli/gateway-cli/run-loop.test.ts -- --reporter=verbose
• Reproduced issue #78463 on Linux 6.8 x86_64 with Node v22.22.0 and OpenClaw 2026.5.5 by seeding a stale running subagent task plus durable child session row; pnpm openclaw tasks audit --json reported stale_running: 1 and errors: 1.
• Re-ran the same seeded-state proof on current OpenClaw 2026.5.6 after this fix; audit reported stale_running: 0, errors: 0, and one lost task with restart-interruption detail.

Real behavior proof

• Behavior or issue addressed: Fixes issue #78463, where openclaw gateway restart could leave active background task records in running forever, causing openclaw tasks audit --json to report stale_running restart leftovers.
• Real environment tested: Linux 6.8 x86_64, Node v22.22.0, OpenClaw 2026.5.5 before fix and OpenClaw 2026.5.6 plus this branch after fix, both with an isolated OPENCLAW_STATE_DIR.
• Exact steps or command run after this patch: seeded the same restart-leftover running subagent task and durable child session row, invoked the restart-interruption marking path, then ran pnpm openclaw tasks audit --json against that isolated state directory.
• Evidence after fix: copied live terminal output from pnpm openclaw tasks audit --json after this patch:

{
  "summary": {
    "total": 1,
    "warnings": 1,
    "errors": 0,
    "byCode": {
      "stale_queued": 0,
      "stale_running": 0,
      "lost": 1,
      "delivery_failed": 0,
      "missing_cleanup": 0,
      "inconsistent_timestamps": 0
    }
  },
  "findings": [
    {
      "severity": "warn",
      "code": "lost",
      "detail": "task interrupted by gateway restart before completion",
      "status": "lost",
      "task": {
        "runtime": "subagent",
        "runId": "run-issue-78463-pr-fixed",
        "status": "lost",
        "error": "task interrupted by gateway restart before completion"
      }
    }
  ]
}

• Observed result after fix: pnpm openclaw tasks audit --json reports stale_running: 0 and errors: 0; the interrupted task is now lost with task interrupted by gateway restart before completion instead of remaining running indefinitely.
• What was not tested: full local pnpm check:changed was not completed because it expanded to broad lanes=all in this environment and Blacksmith/Testbox is unavailable; targeted formatter, diff, regression tests, and real audit reproduction/proof were completed.

Notes

• pnpm check:changed expanded to broad lanes=all in the clean PR worktree and was stopped per repo guidance because Blacksmith/Testbox is unavailable in this environment; targeted touched-surface proof above passed.

Changed files

• src/cli/gateway-cli/lifecycle.runtime.ts (modified, +4/-1)
• src/cli/gateway-cli/run-loop.test.ts (modified, +8/-0)
• src/cli/gateway-cli/run-loop.ts (modified, +11/-0)
• src/tasks/task-registry.maintenance.issue-60299.test.ts (modified, +38/-0)
• src/tasks/task-registry.maintenance.ts (modified, +30/-0)