Gates, hooks and not getting burned

The difference between asking Claude not to do something and making it impossible.

What you'll have at the end

• A hook that blocks writes to secrets
• A hook that blocks destructive commands
• Both failing closed

Chapters

1. Requests versus enforcement — CLAUDE.md is context, not configuration. Anthropic says so plainly. (12 min)
2. Your first blocking hook — Twenty lines of shell that stops the expensive mistake. (25 min)
3. Stop it deleting things — The commands worth blocking, and why archive beats delete. (20 min)
4. Fail closed, not open — A guardrail that fails open is worse than none, because you stop watching. (18 min)

All courses · Start here · Privacy