# AGENTS.md
> Canonical AI policy for this repo.
> If model-specific mirrors exist (`CLAUDE.md`, `GEMINI.md`), generate them from this file.
## Project
[One paragraph: what this project is, what it does, and who it is for.]
## Coding Standards
- Read the highest-signal project-truth docs before deep implementation work.
- Prefer small, reversible changes.
- Keep documentation factual. Tag claims as `[DESIGN]`, `[OBSERVED YYYY-MM-DD]`, or `[CONSTRAINT]` when useful.
- Prefer deterministic scripts over hand-maintained generated files.
## Architecture Rules
- [List the invariants where a violation cascades.]
- Example: no direct database writes outside the data-access layer.
- Example: public API changes require tests and docs in the same task.
## Workflow
- Plan before implementing. Name the files you expect to touch and why.
- Run `git status` and `git diff --minimal` before any commit.
- Stage explicit paths. Never use `git add .`.
- Show the diff to the human and wait for approval before committing.
## Testing
- Run the smallest check that proves the change.
- If you skip a check, say why.
## Safety
- No secrets in docs, code snippets, or committed files.
- Stay within the declared task scope.
- Ask before touching deployment, auth, billing, or data-migration paths.