What problem does it solve?
AI coding agents often make unverified assumptions about codebase dependencies, data schemas, and existing patterns when performing edits, leading to broken code, unintended side effects, and lower quality output that fails to match project conventions.
Core Features & Use Cases
- Pre-Action Fact Gate: Blocks Edit, Write, and Bash commands until concrete investigation facts are gathered, including all files that import the target module, affected public functions/classes, data file schemas and date formats, and the user's exact instruction.
- Multiple Gate Types: Includes specialized gates for first edits per file, new file creation, destructive Bash commands, and routine Bash commands to balance safety and workflow speed.
- Proven Quality Improvement: Independent A/B tests show gated agents produce code with an average +2.25 point higher quality score than ungated agents by forcing context-gathering that eliminates guesswork.
- Use Case: When editing a shared authentication utility used across 12 modules, the gate forces you to first list all importing files to avoid breaking login flows in unrelated parts of the application.
Quick Start
Use the gateguard skill to enable the fact-forcing pre-edit gate for your next code modification task to ensure all dependent modules and data schemas are accounted for before making changes.
