# AGENTS.md — project guide for AI coding agents

Scope: this repository.

Keep this file short and project-specific. Do not put secrets here.

## Project

- Purpose: <one sentence>
- Main stack: <framework/language/runtime>
- Main entry points: <paths>
- Architecture/design docs: <paths if important>

## Read before editing

1. Check `git status` and current branch.
2. Read the files directly relevant to the task.
3. Read deeper/nested `AGENTS.md` or `CLAUDE.md` files when their scope applies.
4. Preserve unrelated uncommitted changes.

## Commands

Use the actual commands for this repository.

```text
Install: <command>
Targeted tests: <command>
Full tests: <command>
Lint/typecheck: <command>
Build: <command>
```

## Coding rules

- Make the smallest correct change.
- Match existing project style and architecture.
- Do not refactor unrelated code.
- Add/update tests for changed behavior when appropriate.
- Do not expose secrets or copy `.env` into logs/chat.
- For bugs: reproduce when possible, then add a regression check.
- For UI/API changes: verify the observable result, not only command exit codes.
- If a command fails, read stdout/stderr and inspect actual state before retrying; fix the smallest cause and do not loop blindly.

## Delivery

- Before finishing: review `git diff` / `show_changes`.
- Do not push, deploy, run lossy migrations or delete production data unless the user explicitly requests that operation.
- If delivery is requested, verify the resulting remote/deployed state.

## Definition of done

The requested behavior exists, relevant checks pass, and remaining unverified items/risks are stated explicitly.