capt-hook 9.20.0

Stop repeating yourself to Claude.

Stop repeating yourself to Claude. captain-hook mines your transcripts for the corrections you keep giving and opens PRs that turn each one into a typed, tested Python hook.

Get started

uvx capt-hook init

init scaffolds .claude/hooks/, wires Claude Code's settings, and arms the session reviewer. One block_command later, a force-push dies at PreToolUse and the hook's inline tests run green:

Driving with an agent? Paste this:

/plugin marketplace add yasyf/captain-hook
/plugin install captain-hook@captain-hook

Prefer a prompt over the plugin?

Run `uvx capt-hook init` in this repo, write one hook that blocks force-pushes,
and verify it with `uvx capt-hook test`. Read https://yasyf.github.io/captain-hook/
if you get stuck.

---

Use cases

Block force-push and rm -rf before they run

One bad Bash call rewrites shared history or eats a directory, and by the time you spot it in the transcript it already ran. Declare the block once, tests inline:

# .claude/hooks/safety.py
from captain_hook import Allow, Block, Input, block_command

block_command(
    ["git", "push", "--force"],
    reason="Force-pushing rewrites shared history",
    hint="Use `git push --force-with-lease` instead",
    tests={
        Input(command="git push --force"): Block(),
        Input(command="git push origin main"): Allow(),
    },
)

The next git push --force never executes: the agent sees BLOCKED: Force-pushing rewrites shared history plus the hint, and reaches for --force-with-lease instead. rm -rf is one more block_command away.

Turn repeated corrections into rules Claude can't forget

You've typed "use uv, not pip" in a dozen sessions, and session thirteen makes the same mistake. After init, the session reviewer reads each transcript as the session ends, keeps the corrections that are standing rules, and — once a pattern proves itself across sessions — opens a PR that codifies it as a hook. Watch the pipeline:

uvx capt-hook status

The dashboard lists every correction it's tracking, staged from first sighting to open PR. You review the PR like any other; merged hooks enforce the rule from then on.

Gate "done" until the tests actually pass

The agent declares victory while the suite is red. A Stop gate holds the line:

# .claude/hooks/quality.py
from captain_hook import RanCommand, TouchedFile, gate

gate(
    "You edited Python files but never ran the tests. Run `uv run pytest` before finishing.",
    only_if=[TouchedFile("**/*.py")],
    skip_if=[RanCommand(r"\bpytest\b")],
)

The agent can't end the turn until a pytest run shows up in the transcript, and the gate stands down on its own once one does.

More in the docs

• Session reviewer — the full corrections lifecycle, from transcript to merged hook PR — guide
• Conditions — typed filters over tools, files, commands, and transcript history — guide
• LLM hooks — gate on a model's verdict when a regex can't decide — guide
• Workflows — multi-step Stop gates with artifact checks and checklists — guide
• Packs — the shipped general, python, and go hook packs — guide
• Testing — run uvx capt-hook test --json in CI so a regressed hook fails the build — guide

Read the docs for the full guide. Licensed under PolyForm Noncommercial 1.0.0, free for noncommercial use.