gateguard
Fact-forcing gate that blocks Edit/Write/Bash (including MultiEdit) and demands concrete investigation (importers, data schemas, user instruction) before allowing the action. Measurably improves output quality by +2.25 points vs ungated agents.
- installs 8w
- 1,717
- 30-day movement
- starts with the next reading
- Related entries
- 3
- Connections
- 0
GateGuard is a PreToolUse hook that blocks Edit, Write, MultiEdit, and Bash actions until the agent produces concrete investigation facts (importers, data schemas, user instructions) rather than self-evaluation. It ships as a zero-install hook in the ECC plugin or as a full package with a .gateguard.yml config, and reports a +2.25 point quality gain in A/B tests.
Self-evaluation prompts don't work; forcing evidence before edits measurably improves agent output.
Use it to
- Gate first edits to shared files
- Block destructive bash commands like rm -rf or git reset --hard
- Require schema checks before writing data files
- Exempt paths via GATEGUARD_EXEMPT_GLOBS
- Customize gate messages per project
For Teams using AI coding agents on multi-module codebases
- Host repository
- affaan-m/ECC
- Installs, lifetime
- 2,900
- Installs, 8 weeks
- 1,717
- Host stars
- 259k
- Host language
- JavaScript