| name | recipe-agent-team-operational-audit |
| description | Recipe: Read-only daemonless audit of agent-team runs, messages, events, stale tasks, unresolved inboxes, and CLI schema. Use for operational inspection, recovery assessment, runtime debugging, trace review, or contract export. Do not use to create runs, dispatch workers, mutate state, plan, design, align terms, compound learnings, or implement. |
| metadata | {"version":"1.1.0","openclaw":{"category":"recipe","domain":"agent-orchestration"},"requires":{"bins":["agent-team"],"skills":["agent-team-shared","agent-team-ops"]}} |
Agent Team Operational Audit
Use this recipe to inspect state, unresolved communication, event history, or the command contract.
Boundary
Use this recipe for audit and diagnosis only. Required runtime skills are declared in metadata; load agent-team-shared first when executing commands. Do not mutate state from this recipe except for a separate user-approved recovery step using the run lifecycle or task command skills.
Steps
- Find active runs.
agent-team run list --status open
- Inspect run status and summary.
agent-team run status --run RUN_ID
agent-team run summary --run RUN_ID
- Inspect stale or unresolved work.
agent-team task stale --run RUN_ID --older-than 2h
- Inspect unresolved run messages.
agent-team message list --run RUN_ID --unread
- Inspect event history.
agent-team event log --run RUN_ID
- Poll incrementally when needed.
agent-team event log --run RUN_ID --after-version STATE_VERSION --limit 100
- Export the CLI contract for tooling or docs.
agent-team schema export
Notes
- Use
event log for auditability, not for replacing task or inbox commands.
- Use
schema export as the command contract source of truth.
- If handing off to run lifecycle for closure, close runs only after all tasks are in a terminal state:
done, failed, or cancelled. Tasks still in_progress, blocked, or pending will cause run_not_ready.
Completion
This recipe is complete when:
- active runs, stale tasks, unresolved messages, and event history relevant to the question have been inspected
- findings cite concrete
run_id, task_id, or msg_id values, not summaries alone
- no state mutation occurred from this recipe
- recovery handoff (if any) names the lifecycle or task recipe that will mutate state, with explicit user approval recorded
