Skip to content

AGENTS.md

Latest commit

 

History

History
51 lines (36 loc) · 2.02 KB

AGENTS.md

File metadata and controls

51 lines (36 loc) · 2.02 KB
Raw

Project Introduction

ai-commit is a Bun-based TypeScript CLI repository for AI-assisted commits.

Information

  • Run commands from the repo root.
  • The package manager used is bun.
  • Primary runtime targets are bun for scripts and node >= 24 for the built CLI.
  • Avoid index.ts barrel files;

Structure

  • .ai-commit/: project-specific support files
  • src/: main source files
  • tests/: automated tests
  • docs/: project documentation

Working Loop

  1. Identify the target area.
  2. Follow local patterns in that directory.
  3. Run focused checks/tests for the target.

TypeScript Preflight Rule

  • Before any TypeScript planning, multi-file edit, refactor, or architecture question, load the architecture-preflight skill first. Re-run it after substantial changes to verify alignment.
  • Trigger on: definitions, types, signatures, exports, services, layers, schemas, wiring, module boundaries, or broad source reading.
  • Posture: inspect declarations/signatures first, then narrow to implementation files only as needed.

Quick Commands

bun run check
bun run build
bun run test
bun run generate:schemas

Learning from reference repositories

  • Check local project support material first when you need examples, prior art, or implementation guidance.
  • Prefer repository documentation and nearby source patterns over browsing generated build output or digging through node_modules/, unless you specifically need implementation-level confirmation.

Engineering Principles

  • Proactive Progress: Don't wait for instructions. Identify blockers, propose solutions, and push work forward autonomously.
  • Robust & Scalable: Prefer solutions that work reliably and can grow. Avoid fragile hacks that break under load.
  • Globally Optimal: Consider the whole system, not just the immediate fix. Trade-offs should be conscious and documented.
  • Verify Reality: Test assumptions. A working demo beats a perfect plan.
  • Ship & Iterate: Perfect is the enemy of done. Get to working state, then improve.