Claude Code Safety Audit

Professional review of your Claude Code setup — by the creator of cc-safe-setup

Is Your Claude Code Setup Safe?

800+ hours of autonomous Claude Code operation. Every mistake documented.

914 example hooks 276 files under tests/ 800+ hours operated

Real Problems We've Seen

File Deletion"Claude ran rm -rf and deleted 3,467 files (7GB)" — #46058
Token WasteCache read rate drops from 89% to 4%, consuming 20x tokens — #40524
Database Destruction"Claude dropped all tables in production" — #46684
Secret LeaksAPI keys committed to public repos or exposed via MCP — #46741
Instruction Ignoring"Wrote it in CLAUDE.md 5 times, still ignored" — #44953
Force PushClaude pushed untested code to main at 3am — real incident from 800h operation

What You Get

Free: Self-Assessment

$0

Run it yourself, get your score instantly:

npx cc-health-check
  • Safety score out of 100
  • Category breakdown (Safety, Quality, Monitoring, Recovery, Autonomy)
  • Specific recommendations with links

cc-health-check on GitHub →

Paid: Professional Audit not accepted

From $50

This is no longer offered. A per-order audit means a person has to take on each one, and that capacity isn't something I can responsibly commit to. Rather than delete the listing, here is what it used to include and what replaced it.

  • Full review of your settings.json, CLAUDE.md, and hook configuration
  • Custom hook recommendations for your specific workflow
  • Token optimization analysis (what's wasting your tokens)
  • Autonomous operation safety assessment
  • Written report with prioritized action items
  • Hook code provided — copy-paste ready

Instead, the free self-audit above covers the same ground mechanically: npx github:yurukusa/cc-safe-setup --audit reports which classes of danger your current setup does and doesn't guard, and --doctor tells you which hooks are misconfigured. For a shared team baseline there is the Team Safety Rollout Pack (¥3,000, one-off, no inquiry needed).

Why This Audit

Claude Code ships with zero safety hooks. The official docs tell you how to write hooks, but not which ones you need. With 914 example hooks tested across 800+ hours of real autonomous operation, we know which 8-15 hooks actually matter for your workflow — and which are noise.

This isn't theory. Every recommendation comes from a real incident. Every hook has been tested in production.

910example hooks
15,795test cases passing
261files under tests/
4GitHub stars
239articles published
12Zenn books

Re-counted on 2026-08-09 against the repository and the platform APIs. Four of these were wrong before: the page said "700+ hooks" in one place and "900+" in another (the actual count of examples/*.sh on main is 910), "9,200+ tests" was corrected badly earlier on this same day: I replaced it with a file count (261 files under tests/) without ever running the suite. Run it and it reports 15,795 test cases, all passing — so the original claim was understated, and my correction of it was wrong. Counting files is not counting tests — and "8 GitHub stars" was overstated (it is 4). Two were understated and are now current: 57 → 239 articles, 3 → 12 Zenn books. A "0 incidents after setup" badge also used to sit here; it has been removed, because I can't substantiate it and this year I have found several of my own guards silently failing.

How It Works

Step 1: Run the Free Assessment

npx cc-health-check

Share your score and output. This tells us your current state.

Step 2: Share Your Setup (Paid)

Send your settings.json and CLAUDE.md (remove any secrets first). Tell us about your workflow: solo or team? Autonomous or interactive? What languages/frameworks?

Step 3: Receive Your Report

A detailed document with:

Contact

Reach out via any channel:

Mention "Safety Audit" in your message. Include your cc-health-check score if you have it.

日本語対応
日本語でのご相談も対応しています。
詳しくは @yurukusa_dev までDMでお気軽にどうぞ。

Learn More