Skip to content

AGENTS.md

Latest commit

 

History

History
69 lines (53 loc) · 2.56 KB

AGENTS.md

File metadata and controls

69 lines (53 loc) · 2.56 KB

AGENTS.md

This is the operating manual for AI agents working in this repository. Read it fully before starting any task. For the project's purpose and goals, read README.md; for what to do next, read TODO.md.

Workflow (this is a methodology-first repo)

  1. Context first. Read README.md and TODO.md, then infer the rest from the existing code and neighboring conventions.
  2. No ping-pong. Don't ask questions one at a time. Resolve unknowns with sensible defaults and announce them. If something is genuinely blocking, ask all such questions in a single batch, then proceed.
  3. Plan then execute. Work through TODO.md plan by plan, chaining features and commits autonomously so the human can do something else meanwhile.
  4. Self-review. After generating code, run the linter/analyzer (see below), fix what it flags, and add or update tests.

Commands

  • Install: <...>
  • Run / dev: <...>
  • Build: <...>
  • Test: <...>
  • Lint / format: <...>
  • Static analysis: SonarQube / SonarScanner (run before closing a plan)

Conventions

  • Language: English for code, identifiers, commits and documentation. Client-facing UI in another language only when end users require it.
  • Code style: follow the existing style and the project's linter/formatter. No superfluous dependencies.
  • Structure: <describe folder layout / entry points>.
  • Docs: keep README.md accurate; add ARCHITECTURE.md when routing or data flows need explaining.

Git

  • Atomic commits with clear English messages.
  • Never add Co-Authored-By: Claude or any AI co-author trailer to commits or PR descriptions.
  • Work on a dedicated branch; do not push to main unless asked.

Quality gate

  • Run static analysis (SonarQube) and the linter after generating code.
  • Create or update tests for new behavior.
  • Don't consider a plan done until analysis is clean and tests pass.

Do NOT

  • Touch confidential/proprietary assets; keep them in a gitignored folder (e.g. data/, datas/) and never commit them.
  • <other project-specific guardrails>

If you also use Claude Code, point its CLAUDE.md at this file (e.g. a single line: See AGENTS.md) so there's one source of truth.