| name | long-running-job-inspector |
| description | Inspect, summarize, cancel, retry, or recover Prometheus background tasks, scheduled jobs, process runs, approvals, and stalled executions. Use when the user asks what is running or needs help with a specific active, stalled, or failed job. |
Long-Running Job Inspector
Use this skill when Prometheus needs to understand or manage ongoing work.
Current State
Status: mostly usable from existing APIs/tools; needs one unified inspector tool for 110%.
Current support:
- Background tasks:
/api/bg-tasks, /api/bg-tasks/:id
- Evidence:
/api/bg-tasks/:id/evidence
- Task stream:
/api/bg-tasks/:id/stream
- Task pause/resume/restart/message routes exist.
- Processes:
/api/processes, /api/processes/:runId, /api/processes/:runId/log, kill/rerun/write/submit.
- Scheduled jobs:
/api/tasks, schedule tool family.
- Task statuses include
running, stalled, needs_assistance, awaiting_user_input, complete, failed.
- Process tools:
process_status, process_log, process_wait, process_kill, process_submit.
Procedure
- List running/stalled/needs-assistance tasks.
- Load the task and evidence bus.
- Inspect latest journal entries, last tool call, pause reason, and runtime progress.
- If process-bound, inspect process list and logs.
- If approval-bound, inspect pending approval state.
- Recommend one action: wait, nudge/message, resume, retry, cancel/kill, or ask user.
Gap To Implement
Add job_inspector as a single read-only summary tool with optional explicit follow-up actions.