capt-hook 3.3.2

Declarative hook framework for Claude Code

captain-hook

Declarative hook framework for Claude Code. Write hooks as data, test them inline, and ship them to CI in the same shape they run in production.

Quickstart

No install step — everything runs through uvx. Pick a front door:

From your terminal:

uvx capt-hook init

From inside Claude Code — install the plugin, then ask Claude to set it up:

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

set up captain hook

Either path lands in the same place: .claude/hooks/ scaffolded, Claude Code's settings wired, the bundled skills installed, and the session reviewer watching this repo. uvx fetches captain-hook into a throwaway environment, so it never enters your pyproject.toml — and every command below works the same way once you prefix it with uvx.

Your first hook

A hook is declarative Python with an event, some conditions, and an action. This one stops the agent from finishing a UI change it never looked at.

# .claude/hooks/visual_review.py
from captain_hook import gate, TouchedFile, UsedSkill

# A Stop gate: before the agent finishes, block if it edited UI files without doing a visual review.
gate(
    # the one-line reason shown to the agent when the gate fires
    "You edited UI files. Open them with agent-browser and verify they render before finishing.",
    # fires only if UI files changed
    only_if=[TouchedFile("**/src/routes/**", "**/src/components/**")],
    # already reviewed -> don't block
    skip_if=[UsedSkill("agent-browser")],
)

Conditions match tools, files, commands, and even which skills the agent used.

Test your hooks

Every deterministic hook carries inline tests, so a broken hook fails like broken code. Run them from your project root, where --hooks defaults to .claude/hooks.

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

block_command(
    ["git", "stash"],
    reason="Use the team's VCS workflow for shelving changes",
    hint="Commit a WIP change instead of stashing",
    tests={
        Input(command="git stash"): Block(),
        Input(command="git status"): Allow(),
    },
)

uvx capt-hook test

init already wired Claude Code's settings. Each event runs uvx capt-hook run <Event>, with the event JSON arriving on stdin and the verdict written to stdout. Re-run uvx capt-hook register-hooks only after you add hooks on a new event; it writes .claude/settings.local.json for you.

Session reviewer

init also turns on the session reviewer. When a Claude Code session ends, it mines the transcript for the durable corrections you gave and the hooks that misfired, judges each one, and — once a pattern clears its thresholds — opens a pull request that adds a new hook or fixes the one that misfired. You review the PR like any other.

It's on by default after init. Turn it off for a repo with uvx capt-hook review disable, or skip it at setup with uvx capt-hook init --no-review. The session reviewer guide covers prerequisites (an authenticated claude and gh) and the HOOKS_REVIEW_* tuning knobs.

Agent Skills

captain-hook ships two Agent Skills so you don't have to write hooks by hand. bootstrapping-hooks surveys your repo's docs, CI, and git history and proposes gates and nudges; translating-styleguides turns a STYLEGUIDE.md into enforced rules. Both land in .claude/skills/ via init and ship as the plugin in the Quickstart — ask Claude to "set up captain hook" and bootstrapping-hooks takes it from there.

What this solves

captain-hook covers these jobs:

• Block dangerous tool calls before they execute on PreToolUse, like force-push, package-manager footguns, and raw rm -rf.
• Drive the agent with feedback that fires on the patterns it actually emits, such as repeated failures, weakened tests, and missed conventions.
• Enforce multi-step workflows with Stop gates and artifact validation, so the agent can't declare "done" without running tests, writing a report, or completing a checklist.
• Keep all of the above testable. Every hook ships with inline tests = {...} that uvx capt-hook test runs in CI, so you catch broken hooks the way you catch broken code.

Docs

Read the docs for the full guide to conditions, primitives, LLM hooks, workflows, state, and real-world patterns.

For working on captain-hook itself, see the development guide.

License

Licensed under PolyForm Noncommercial 1.0.0, free for noncommercial use.