What problem does it solve? Claude Code hooks fail silently in confusing ways: a hook that never fires looks identical to one that fires and finds nothing, exit code 1 does not block, and matchers switch between exact strings and regexes based on their characters. This Skill provides a verified reference covering all 31 hook events and five handler types, plus a drift detector that checks the documentation against upstream docs before you rely on it. ## Core Features & Use Cases - Complete hook reference: All 31 events, matcher semantics, the five handler types (command, http, mcp_tool, prompt, agent), exit-code behavior per event, and JSON output contracts, organized into four reference files. - Freshness verification: A bundled script compares the shipped reference against upstream Claude Code docs and reports CURRENT, COSMETIC, STRUCTURAL, or BROKEN verdicts, including whether your local CLI is newer than the documented behavior. - Authoring guidance: Traps that bite hardest (fail-open crashes, exit-0 stderr invisibility, PreToolUse reading unapproved files), testing hooks from stdin without a live session, and measuring a guard's precision before escalating a nudge to a block. - Use Case: You want a hook that blocks dangerous Bash commands. The Skill tells you PreToolUse is the right event, that exit 2 blocks while exit 1 does not, that your matcher may be an unanchored regex, and how to test the whole thing by piping JSON on stdin. ## Quick Start Ask the AI to help you write a PreToolUse hook that blocks a specific Bash command pattern, and have it run the freshness check first.
