Free · one email · 6-part playbook
The Claude Code Field Guide
How I run Claude Code on two real AI systems every day: the onboarding file, plans that outlive the chat, agents with one job, guardrails in code, gates before done, and the model bill. Six parts, each with a checklist.
What’s inside
- 1The onboarding fileOne doc the agent reads before it touches anything: what this is, where things live, what must never break.
- 2Plans that outlive the chatThe plan lives in the repo as an index plus one small file per chapter, so any session can pick it up.
- 3Agents with one jobSubagents with a narrow brief, skills for repeatable procedures, lanes for what each may change, and a limit on how many run at once.
- 4Guardrails in code, not in the promptPropose-only tools, permission deny-lists, hooks that block, and guard scripts that fail the build.
- 5Done means checkedGates that run before every commit, chained so a failure stops it, and the ways a green result can lie.
- 6Models, money and limitsThe right model per job, measured; a ledger of every call; regeneration driven by inputs, never a clock.
Already subscribed? Use the same email and I'll send you straight in.