tidy-skill

Case: tidy-skill repository self-audit

Kind: real product repository
Root: this repo (Phoenix0531-sudo/tidy-skill)
Scripts: own portable auditors
Snapshot: 2026-08-05 author-run (paths redacted in stored reports)

Before portfolio hardening (historical)

Early public docs state was effectively:

• Root README pair acted as thin placeholders while the long English body lived in a side file
• No committed self-audit reports under docs/self-audit/
• No banner / terminal preview assets
• Missing tracked .agent_tmp/ / .agent_reports/ layout markers → hygiene score sat around 77 / 100 on author runs

After hardening + P0–P2 upgrades

Report	Result
Repo hygiene score	100 / 100 — Clean
Agent artifact audit	0 suspicious root files
Dev environment audit	90 / 100 — Highly controlled
Doctor	Package + hygiene pass (v1.4.0)
Fixture evals	8 / 8 PASS

Additional product surface now present:

• Bilingual portfolio README
• npx skills add Phoenix0531-sudo/tidy-skill --skill tidy-skill verified discoverable
• Optional host samples under docs/host-samples/
• Per-platform install notes under docs/platforms/
• v1.5.0: planning-with-files coexistence (planning_root_globs, .planning/), peer comparison doc
• v1.4.0: project policy, doctor, classify, score snapshots/gate

What changed operationally

1. Process files stay out of root (rule templates + skill guidance).
2. Layout directories exist with .gitkeep and are not counted as agent litter.
3. Self-audit reports are redacted and committed as evidence, with methodology footnotes.
4. Cleanup remains DryRun-first; no auto-destructive environment operations.

Limits

• Self-audit by own scripts, not an independent lab.
• Dev-environment score reflects one Windows author machine.
• Marketplace plugin listings for Claude/Codex are still not claimed.

This site is open source. Improve this page.