The Claude Code Operator's Handbook

You gave an AI agent shell access to your machine. This is the operations manual you should have read first — 18 chapters, ~14,000 words, on running coding agents safely and efficiently in real work. Not prompt tips; operational practice.

Get the handbook — $29 Read 3 chapters free

What's inside

Part I — The Threat Model

1. How agents actually fail (blast radius, not malice)
2. The permission model: allow, deny, or gate
3. Secrets, keys, and the .env that almost shipped

Part II — Guardrails

4. Hooks deep-dive: PreToolUse, exit codes, fail-closed design
5. Dangerous-command gating patterns
6. Multi-session and worktree safety
7. Pre-push secret scanning that actually works

Part III — Cost & Efficiency

8. Where tokens really go
9. Prompt-cache mechanics and hit-rate tuning
10. CLAUDE.md as a cost lever
11. Tool-output discipline

Part IV — Workflows & Recovery

12. The /ship pipeline
13. Onboarding a repo (and an agent) fast
14. Team conventions without drift
15. Recovery playbooks
16. Auditing what an agent did
17. Scaling to parallel agents
18. A maturity model for agent operations

Who it's for

• Developers running Claude Code (or any coding agent) daily on real codebases.
• Teams adopting agents who need shared, safe operating practice.
• Anyone who's had — or wants to avoid — the "the agent did *what*?" moment.