返回 CodeWhale
SKILL.md
1 ---
2 name: fleet-manager
3 description: Use when managing, triaging, restarting, escalating, or summarizing Codewhale Agent Fleet runs and workers.
4 metadata:
5 short-description: Triage Codewhale Agent Fleet runs
6 ---
7
8 # Fleet Manager
9
10 Use this skill when acting as a manager agent for Codewhale Agent Fleet runs.
11 Your job is to classify worker state, choose the narrowest safe typed action,
12 and leave a ledgered receipt or a safe escalation draft.
13
14 ## Authority Boundary
15
16 - Prefer typed fleet surfaces over shell spelunking: `codewhale fleet status`,
17 `inspect`, `logs`, `artifacts`, `interrupt`, `restart`, `stop`, and the
18 Runtime API fleet endpoints.
19 - Do not read `.codewhale/fleet.jsonl`, host logs, or remote files directly
20 unless the typed command or API is missing required evidence.
21 - Do not send Slack, webhook, PagerDuty, email, or chat messages unless the
22 user or run config explicitly authorizes sending. Draft the message instead.
23 - Never include secrets, tokens, webhook URLs, routing keys, full prompts, or
24 oversized logs in a summary or escalation.
25
26 ## Triage Loop
27
28 1. Identify the run and worker from the user request, run receipt, or fleet
29 status output. If no worker is named, start with `codewhale fleet status`.
30 2. Inspect the worker with `codewhale fleet inspect <worker-id>` or the matching
31 Runtime API worker endpoint.
32 3. Review bounded evidence with `codewhale fleet logs <worker-id>` and
33 `codewhale fleet artifacts <worker-id>`. Summarize artifact refs, not full
34 payloads.
35 4. Classify the state before acting:
36 - `transient failure`: transport error, timeout, stale heartbeat, host
37 unavailable, or retryable provider/network failure.
38 - `task failure`: worker completed the task but the result is wrong,
39 missing required artifacts, or reports a domain error.
40 - `verifier failure`: scorer/verifier failed or disagrees with the worker
41 result.
42 - `needs-human`: missing authority, unsafe secret boundary, destructive
43 action, repeated restart exhaustion, ambiguous product decision, or
44 conflict between artifacts and verifier.
45 5. Choose one typed action:
46 - transient and retry budget remains: `codewhale fleet restart <worker-id>`.
47 - transient but unsafe to retry: draft escalation and mark needs-human.
48 - task failure: preserve artifacts, summarize the failure, and avoid restart
49 unless the task spec says retrying can produce new evidence.
50 - verifier failure: inspect scorer inputs and artifacts, then escalate if the
51 verifier cannot be corrected through a typed action.
52 - needs-human: do not restart automatically; draft a concise escalation.
53 6. Record the result in the response: classification, action taken or drafted,
54 evidence commands, artifact refs, and next owner.
55
56 ## Restart vs Escalate
57
58 Restart only when all of these are true:
59
60 - the failure is likely transient,
61 - the task is idempotent or the run policy allows retry,
62 - retry budget remains,
63 - no secret, permission, or destructive action boundary is involved, and
64 - the previous attempt produced enough receipt data to explain the restart.
65
66 Escalate when any of these are true:
67
68 - restart budget is exhausted,
69 - the worker requests secrets or new authority,
70 - artifacts indicate data loss, corruption, or destructive side effects,
71 - the verifier and task result conflict in a way you cannot resolve from typed
72 evidence,
73 - the same failure repeats after a restart, or
74 - a human product or release decision is required.
75
76 ## Safe Escalation Draft
77
78 Use this shape for Slack/PagerDuty drafts. Keep logs to three short lines or an
79 artifact ref.
80
81 ```text
82 Codewhale fleet needs attention
83 Run: <run-id>
84 Worker: <worker-id>
85 Task: <task-id or unknown>
86 Classification: <transient failure | task failure | verifier failure | needs-human>
87 Reason: <one sentence, no secrets>
88 Latest typed evidence: codewhale fleet inspect <worker-id>; codewhale fleet artifacts <worker-id>
89 Safe log excerpt: <3 lines max or "see artifact <ref>">
90 Requested decision: <restart approval | verifier review | task owner review | permission decision>
91 ```
92
93 ## Post-Run Receipt
94
95 End every fleet-manager response with a compact receipt:
96
97 ```text
98 Fleet receipt
99 Run: <run-id>
100 Workers checked: <count/list>
101 Classification: <state>
102 Action: <restart/interrupt/stop/escalation draft/no-op>
103 Ledger expectation: <typed action should be recorded | draft only, no send>
104 Artifacts reviewed: <refs>
105 Follow-up owner: <manager | task owner | human>
106 ```
107
107 lines MARKDOWN