AGENT SKILL

log-hygiene

暂无简介

License 许可
MIT
最近更新
2026-06-08
Stars
★ 0
Forks
0

03 SKILL.md 正文

log-hygiene

when_to_use

Use this skill when inspecting long-running detached agent runs, large logs, failed validation, or any workflow that might tempt a maintainer to paste logs into chat.

inputs

  • Task id.
  • .run marker path.
  • Log path.
  • Done marker path.
  • Completion JSON path.
  • Log-size cap.

allowed_actions

  • Show active process status.
  • Show done marker presence.
  • Show log size only.
  • Show completion JSON readiness.
  • Use narrow status/token grep.
  • Record raw log paths in receipts.

forbidden_actions

  • Do not run cat "$LOG".
  • Do not run broad tail -200 "$LOG".
  • Do not paste full logs into docs, prompts, receipts, or chat.
  • Do not dump full diffs as a substitute for a receipt.

commands

scripts/wave peek --task <task>
wc -c /tmp/<task>_<timestamp>.log
grep -Eim 5 "token|status|error|failed|passed|complete" /tmp/<task>_<timestamp>.log || true

stop_conditions

  • Log exceeds the configured cap and the current task is broad exploration.
  • The .run marker is missing.
  • The done marker and active process disagree.
  • Completion JSON is expected but missing.

completion_receipt_rules

  • Include log path and log size.
  • Include only summarized error/status lines.
  • Include the command used to inspect status.
  • Include next action when log inspection finds a blocker.

04 同仓库更多 Skills