# AGENTS.md

> Auto-generated by `npx @jamie-tam/forge init`. Run `/setup` in Claude Code to detect your stack and fill in the project profile.

## Project

<!-- Filled by /setup after stack detection -->
- **Name:** (run /setup)
- **Stack:** (run /setup)
- **Test runner:** (run /setup)
- **Package manager:** (run /setup)

<!-- FORGE:START — do not edit between markers; updated by forge -->

> This file is imported by `.claude/CLAUDE.md` via `@AGENTS.md` so Claude Code reads it. It also serves as the ecosystem-standard agent-context file (read by Cursor, Codex, and other tools). For the forge command catalog, skill prefix groups, and `.forge/` directory layout, see `.claude/CLAUDE.md`.

## Required Checks

- Run tests before committing
- Every commit must leave the project buildable
- No `--no-verify` on git commands
- No hardcoded secrets

## Commit Conventions

Format: `type(scope): description` — lowercase, imperative mood.
Types: `feat`, `fix`, `refactor`, `test`, `docs`, `chore`.

## Key Constraints

- Quality gates block phase transitions until criteria are met (see `.claude/references/common/quality-gates.md`)
- Test-driven development is mandatory for production code (RED → GREEN → REFACTOR)
- Code review is a chain of focused subagents: safety → craft → runtime reachability → gotcha-hunter (→ optional Codex adversarial)
- Work state lives in `.forge/work/{type}/{name}/manifest.yaml` (type: feature|bugfix|refactor|hotfix|greenfield)

## Deeper Docs

| File | Contents |
|------|----------|
| `.claude/CLAUDE.md` | Claude Code project configuration, command catalog, First Prompts |
| `.claude/references/common/quality-gates.md` | Gate definitions and pass/fail criteria |
| `.claude/rules/common/` | Always-loaded coding standards (security, verification, guardrails, skill-selection) |

<!-- FORGE:END -->

## Project-Specific Notes

<!-- Add project-specific context: key decisions, architecture notes, known issues -->