jleechanorg/disk_magician/hermes

Disk Magician — Hermes Skill

This skill teaches the Hermes agent how to manage disk usage, diagnose growth regressions, and execute cleanup tasks.

Commands for Hermes

• Validate System Health:

``bash ./disk_magician.sh alert ``

• Audit candidates for cleanup:

``bash ./disk_magician.sh audit ``

• Perform safe caches and temporary files deletion:

``bash ./disk_magician.sh clean ``

• Review history and regressions:

``bash ./disk_magician.sh history ``

Hermes Guardrails & Diagnostic Recipes

• Ensure you perform a dry-run check before executing any deletions.
• If deleting directories modified within the last 7 days, verify that user approval is in context or prompt the user for confirmation.
• Worktree 7-day rule (single source of truth): see repo CLAUDE.md section "Worktree 7-day rule". Recency comes from scripts/lib/worktree_recency.sh; never re-derive it from stat <wt>/.git or stat <wt> (both measure worktree creation, not use), and treat unmeasurable as protected.
• Worktree Removal Forensics: Before concluding that a missing worktree was cleanly removed with git worktree remove, inspect ls -la $(git rev-parse --git-dir)/worktrees/<name>/ and .git/worktrees/<name>/gitdir. Admin directory presence proves the working tree was deleted manually via rm -rf, not clean git worktree remove.
• Multi-Agent History Search: When searching /history or investigating agent actions, search across Antigravity (~/.gemini), Cursor (~/.cursor), Hermes (~/.hermes), Claude (~/.claude), and Codex (~/.codex) in parallel (see ~/.claude/skills/history-search/SKILL.md).
• mem0 & Qdrant Diagnosis Recipe (Launcher vs API Key):

When mem0 unavailable or Qdrant connection errors occur, do NOT assume an embedder API key is missing (mem0 uses local FastEmbed + Ollama). Follow the 4-step recipe:

1. Probe reachability: lsof -nP -iTCP:6333 -sTCP:LISTEN and curl -sS -m 3 http://127.0.0.1:6333/healthz
2. Inspect launchd: launchctl print gui/$(id -u)/ai.hermes.qdrant and tail -30 ~/Library/Logs/ai.hermes.qdrant.err.log (check for Docker wait failure no usable Docker context after 60s; fix with native /Users/jleechan/.local/bin/qdrant binary, WorkingDirectory, KeepAlive=true, and ProcessType: Interactive).
3. Inspect cwd/storage permissions: Check for Failed to create snapshots temp directory / Read-only file system (os error 30). Fix with explicit WorkingDirectory and absolute paths in config.yaml.
4. Helper API compatibility: mem0 2.x migration requires filters={'user_id': ...} instead of user_id= for m.search() and m.get_all().