richfrem/create-hook

Follow the create-hook skill workflow to design and generate a hook configuration.

Inputs

• $ARGUMENTS — optional hook event type (e.g. PreToolUse, Stop, PermissionRequest) or a use-case description (e.g. "block dangerous bash commands"). Omit for discovery.

Steps

1. If $ARGUMENTS names an event or use case, use it to seed Phase 1 questions
2. Follow the create-hook phased workflow: select event, choose handler type (command / prompt / agent), design the matcher and logic, then write the hook entry
3. Validate with validate_hook_schema.py and test with test_hook.py
4. Report placement (global hooks.json vs skill-scoped frontmatter) and next steps

Output

hooks.json entry or SKILL.md frontmatter block with complete hook configuration (event, matcher, handler type, command/prompt body, output schema).

Hook Script Standards

When generating a Python hook script, always apply these two rules:

Cross-platform python command — use python3 ... || python ... in hooks.json so the hook works on both macOS/Linux (python3) and Windows (python):

{ "type": "command", "command": "python3 ${CLAUDE_PLUGIN_ROOT}/hooks/script.py || python ${CLAUDE_PLUGIN_ROOT}/hooks/script.py" }

Project-type guard — hooks run in every project, not just ones that have initialized this plugin. Add an early-exit guard at the top of main() so the script skips silently in projects that lack the required context:

def main():
    project_root = Path(os.environ.get("CLAUDE_PROJECT_DIR", os.getcwd()))
    if not (project_root / "context").exists():
        return  # Not an initialized project — skip silently
    ...

Adapt the guard to whatever directory/file your hook requires (e.g. .agent/, context/os-state.json).

Edge Cases

• If $ARGUMENTS is empty: begin with the event selection question in Phase 1
• If the requested event is not in the 13 supported events: explain valid options
• If the use case implies a skill-scoped hook (enforce invariant only during one skill): generate frontmatter syntax instead of a global hooks.json entry
• If user wants to auto-approve subagent permissions: use PermissionRequest + prompt handler