Create Hook: Turn an Idea Into a Working Claude Code Hook

What the user wants the hook to do: $ARGUMENTS

If that is filled in, treat it as the behavior spec and start from it. Don't re-ask what they already told you; only ask to pin down the gaps (the exact paths / commands / patterns, and whether it must block). If it is blank, start by asking what the hook should guarantee or do (Workflow step 1).

What a hook is (30-second intro)

A hook is deterministic code that fires automatically on a Claude Code lifecycle event - before a tool runs, after it runs, when the agent tries to finish, when a session starts, and so on. Unlike a skill or a subagent, the agent does not choose to invoke a hook - it fires whether the model "remembers" or not. That is the whole point: a rule asks the agent to behave; a hook guarantees it, at the tooling layer the model can't talk its way around.

The user brings the idea ("the agent should never read my .env"); this skill writes the code and wires it in. They don't need to know Python.