A checklist skill for safer, cheaper AI coding agents
coding-posture is a small tool that makes an choose a work mode before starting a non-trivial task. The modes include debug, fix, review, , refactor, optimize, migrate, upgrade, integrate, spike, and unstuck. Each mode gives the agent a short checklist, so it is less likely to skip reproducing a bug, pretend a test passed, or change without a way back.
Every mode shares basic rules: run the real check, do not weaken a test just to make it pass, and do not use without a clear scope. The design is based on research claiming that a concrete process changes model behavior more reliably than telling the model to act like a certain kind of expert. The includes a small comparison with and without the skill, using an LLM judge and a baseline.
The early result was 85% versus 70% on one model across 5 cases, a 15 percentage point gain. That is only a directional signal because the test set is small, but the can be run from the eval folder.
Key points
- The tool asks an to pick a mode before complex work.
- Modes come with short checklists instead of role-play prompts.
- Shared rules require real , protect tests, and limit .
- The first showed 85% versus 70% on one model over 5 cases.
- The approach may reduce wasted tokens by cutting loops and skipped steps.