AGENT SKILL

orch-weekly-auto-fix

After the weekly opportunity scan identifies issues, run auto-fix patterns — git pull, branch cleanup, Docker restart, permission fixes, disk cleanup — then verify each fix succeeded. Reduces manual intervention for known recurring problems.

License 许可
MIT
最近更新
2026-11-14
Stars
★ 4
Forks
1

01 前言 Front Matter

name
orch-weekly-auto-fix
author
Steadfaste Fleet
license
MIT
version
1.1.0
metadata
{ "hermes": { "tags": [ "cron", "automation", "maintenance", "git", "docker", "security" ], "related_skills": [ "cron-engineering", "public-contribution" ] } }
description
After the weekly opportunity scan identifies issues, run auto-fix patterns — git pull, branch cleanup, Docker restart, permission fixes, disk cleanup — then verify each fix succeeded. Reduces manual intervention for known recurring problems.

03 SKILL.md 正文

Weekly Auto-Fix (with Verification)

When to Use

Load this skill when:

  • Setting up or modifying the weekly opportunity scan cron job
  • A cron agent needs to auto-fix issues it discovers
  • You want the scan-fix-verify-report pattern instead of scan-only

How It Works

The cron runs in four phases:

[Phase 1: Scan]  ← LLM-driven, finds issues across git/docker/perms/disk
    ↓
[Phase 2: Fix]   ← Agent attempts fixes, then runs companion script
    ↓
[Phase 3: Verify] ← Re-check each condition post-fix; PASS/FAIL/WARN per check
    ↓
[Phase 4: Report] ← Compact summary: fixed & verified vs needs attention

Setup

1. Deploy the companion script

Copy scripts/orch-weekly-auto-fix.py from the steadfaste-fleet repo to ~/.hermes/scripts/:

cp steadfaste-fleet/scripts/orch-weekly-auto-fix.py ~/.hermes/scripts/orch-weekly-auto-fix.py

The script handles known fix patterns with built-in verification:

  • git-pull — pulls upstream changes, then verifies behind count is 0 and no conflicts
  • branch-delete — removes stale local branches, then verifies they're gone from git branch
  • permissions — chmod 600/644, then stats the file to confirm
  • disk-cleanup — cleans bun cache if >100MB, then verifies dir is gone or under threshold
  • docker-restart — restarts unhealthy/restarting containers, then checks docker ps for healthy status

2. Create or update the weekly scan cron

hermes cron update <job-id> \
  --prompt "$(cat << 'PROMPT'
You are Moses. This is your weekly opportunity scan with auto-fix and verification.

## Phase 1: Scan

Check repos, Docker containers, system files, and cron jobs for issues:
- Git status (behind origin/main? conflicts?)
- Docker container health (unhealthy/restarting?)
- File permissions (world-readable outputs, executable pids?)
- Disk usage (large caches?)
- Cron job errors

## Phase 2: Auto-Fix

Attempt fixes for each issue found:

**Git fixes:** `git pull --rebase --autostash` if behind. Delete stale branches. Merge ready PRs.

**Docker fixes:** `docker restart <name>` for unhealthy containers.

**Permission fixes:** `chmod 600` / `chmod 644`.

**Companion script:** `python3 ~/.hermes/scripts/orch-weekly-auto-fix.py --verbose` as safety net.

## Phase 3: Verify

For each fix, verify the fix actually worked:
- Git: behind count is 0, no conflict markers in git status
- Docker: container shows "Up" or "healthy" in docker ps
- Permissions: stat confirms correct mode
- Disk: cache dir is removed or under threshold

## Phase 4: Report

- ✅ FIXED if fix + verification passed
- ❌ FAILED if fix or verification failed
- Exit silently if nothing needed doing
PROMPT
)" \
  --schedule "0 8 * * 1" \
  --deliver origin

Fix & Verify Patterns

Pattern Detection Fix Verification
Git behind upstream git rev-list --left-right --count HEAD...origin/main git pull --rebase --autostash Re-check behind count = 0, no UU/AA/DD in git status --short
Stale local branches git branch -vv (detect : gone]) git branch -D <name> Branch absent from git branch output
Docker unhealthy docker ps (status "unhealthy"/"restarting") docker restart <name> docker ps shows "Up" or "healthy" after 3s wait
World-readable scorer stat -f %p shows 644 chmod 600 stat shows 600
Executable pid file stat -f %p shows 755 chmod 644 stat shows 644 or 600
Large bun cache du -sh ~/.bun/install/cache >100MB rm -rf Dir gone or <100MB remaining

Companion Script Output

The script produces structured JSON with verification results:

{
  "actions_taken": ["git-pull: pulled 3 commit(s) from origin/main"],
  "verify_results": [
    {"check": "git-pull", "status": "PASS", "detail": "up to date, no conflicts"},
    {"check": "perms-scorer", "status": "PASS", "detail": "already 600"},
    {"check": "perms-pid", "status": "FAIL", "detail": "mode is 0o755"}
  ],
  "summary": {"fixed": 1, "passed": 2, "failed": 1, "warned": 0}
}

On success with nothing to do, outputs empty JSON {} (silent — watchdog pattern).

Pitfalls

  • Don't merge PRs during a rebase conflict. If git pull --rebase fails, report it — don't force push.
  • Don't delete protected branches (main, master, develop). The script skips them.
  • Docker restart may cause brief downtime. Only restart containers clearly marked "unhealthy" or "restarting" — not healthy ones.
  • The companion script is a safety net, not the primary fixer. The LLM agent should attempt reasoning-heavy fixes (git merge, mycortex repair) directly, then use the script for mechanical checks.
  • If nothing was broken, stay silent. The watchdog pattern applies — the user should not hear from a cron that found nothing to do.
  • Verification adds latency. Docker verification waits 3s; git verification runs 2 extra commands. This is fine for a weekly cron, but don't add heavy verifications to high-frequency crons.

04 同仓库更多 Skills