Free tool

Claude Code hooks builder

Tick the guardrails you want, get a valid .claude/settings.json. No signup, nothing stored.

Block destructive shell commands
Refuses rm -rf, git push --force, git reset --hard and history rewrites before they run. Exit code 2 blocks the tool call and tells Claude why.

Protect .env and credential files
Blocks reads and edits of .env, *.pem, id_rsa, credentials.json so secrets never enter the transcript.

Auto-format every file Claude edits
Runs your formatter on the exact file after each Edit/Write, so you never review a diff full of whitespace noise.

Run the test suite when Claude finishes
Fires on Stop. You find out the change broke something in the same breath Claude says it's done.

Log every shell command to a file
Appends each command plus a timestamp to .claude/bash-history.log. The audit trail you'll want the one time something goes sideways.

Terminal bell when Claude needs you
Fires on Notification, so a long run in a background tab actually gets your attention.

Your stack

Sets the formatter and test command used by the recipes above.

Custom test command (optional)

Where this file goes

your-repo/
└── .claude/
    └── settings.json   <- paste here (checked in, shared with your team)

Use .claude/settings.local.json instead for hooks you don't want to commit. Both are merged; local wins.

What a hook actually is

A shell command Claude Code runs at a fixed point in the loop. It receives a JSON blob on stdin describing the event — which tool, which arguments — and its exit code decides what happens next:

exit 0 — allow, carry on.
exit 2 — block the tool call. Whatever the hook wrote to stderr is shown to Claude, so it can correct itself.
any other code — non-blocking error, logged and ignored.

That exit-2 path is the whole point: it is the only way to make a rule Claude cannot talk itself out of. A line in CLAUDE.md is a request; a hook is enforcement.

The events you can hook

PreToolUse — before a tool runs. The only place you can block.
PostToolUse — after it succeeds. Formatters, linters, type checks.
UserPromptSubmit — on every message you send. Inject context.
Stop / SubagentStop — when the turn ends. Test suites, notifications.
SessionStart / SessionEnd — bookend a session.
Notification — Claude is waiting on you.
PreCompact — before context gets summarized.

The matcher field is a regex against the tool name, so "Edit|Write" catches both and an absent matcher catches everything.

These recipes are the safe subset. The $29 pack ships three complete presets — TDD guard, CI agent, safe defaults — plus a walkthrough of what each event actually fires on, five subagents and 30 workflow recipes.

Free tools: CLAUDE.md generator · Hooks / settings.json builder · Free 4-file sample
Claude Code Power Pack v1.0 · Home · I already paid
Not affiliated with Anthropic. “Claude” and “Claude Code” are trademarks of Anthropic, PBC.
Questions or refund: [email protected]