Gateguard

This is a pre-action hook that blocks Claude's first edit or file write and forces it to gather concrete facts first: what imports this file, what the data schema actually looks like, what the user literally asked for. The claim is that investigation beats self-evaluation, backed by two A/B tests showing a 2.25 point quality jump. MultiEdit gets gated per file, destructive bash commands get gated every time, routine commands once per session. The hook is included and you flip it on via hooks.json, or install the full package for per-project config. If you work on codebases where Claude guesses at date formats or misses downstream callers, this adds friction in exactly the right place.

Install to Claude Code

npx -y skills add affaan-m/ecc --skill gateguard --agent claude-code