Before you start
How to control what Claude Code is allowed to do
This tutorial
What you’ll understand: cckit asks before it changes anything in a repo it hasn’t worked in before. This tutorial walks the permission gate — how consent is recorded, what always runs regardless, and what needs a separate confirmation — so you know exactly what an agent driving cckit can and can’t do on your behalf.
-
Start read-only. In an unknown repo, cckit is read-only by default. Point it at the project and inspect state without granting anything:
Terminal window cckit scan --llm # detect repo root, stack hints, and kit state — reads only -
Consent is recorded before the first mutation. The first time cckit is about to change something, it records your yes in
.cckit/consent(gitignored) — so consent is explicit, local, and never committed. -
Destructive operations still ask every time. Force-push, repo creation, and history rewrites always require a separate, explicit confirmation — consent to mutate does not cover them.
How it works
Section titled “How it works”cckit resolves its config by walking up from where you invoke it (nearest wins: $KIT_CONFIG, then
cckit.config.json, then .claude/kit.config.json, then its own fallback) — no org, repo, or path is
hardcoded. Everything it writes about you — consent, identity, your privacy denylist — lands in
.cckit/, which is gitignored, so it never leaves your machine. Two things are not negotiable:
the secret and privacy guard always runs before anything is committed, and consent never bypasses it.
Troubleshooting
Section titled “Troubleshooting”cckit won’t mutate the repo. It hasn’t recorded consent yet. The first mutating operation prompts
for it; approve once and it’s stored in .cckit/consent.
A destructive op keeps asking. By design — force-push, repo create, and history rewrite always require explicit confirmation, even after you’ve consented to ordinary mutations.
The guard blocked a commit even though I consented. Consent doesn’t bypass the secret/privacy guard — that always runs. See stop Claude committing secrets.
Related tutorials
Section titled “Related tutorials” Stop Claude Code from committing secrets The guard that runs regardless of consent.
Keep private & company data out of AI commits Your local denylist, stored in .cckit/.
Run checks automatically with cckit hooks The guards that fire before a command runs.
Independent, educational project — not affiliated with or endorsed by Anthropic. Claude and Claude Code are trademarks of Anthropic PBC. Disclaimer & trademarks
From Mexico with love by josegtz