Gateguard

Forces Claude to gather concrete facts before editing files or running destructive commands. Instead of "are you sure?" self-evaluation (which never works), it blocks the edit and demands specific investigation: list all importers via grep, show the data schema, quote the actual user instruction. Two A/B tests show a +2.25 point quality improvement because the act of investigation creates context that changes what Claude writes. The trade-off is real: every first edit per file gets blocked once, which slows you down but catches the "I assumed ISO dates when your CSVs use %Y/%m/%d" class of bugs. Worth it on codebases where edits ripple across modules.

Install to Claude Code

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