# AGENTS.md

This project was initialized with the ForgeAI agentic harness (`.ai/`).

At the start of the first agent session, check whether the installed
ForgeAI harness is behind the latest package:

```bash
npx forgeai-agentic-init@latest --check-updates --check
```

If the check reports an outdated harness, ask the human whether to skip for
now or update to the latest version before implementation work. If the human
chooses update, the agent runs:

```bash
npx forgeai-agentic-init@latest --upgrade
```

Before making any code changes, read these files in order:

1. `.ai/README.md` — harness overview and full recommended read order
2. `.ai/PROJECT.md` — project identity, stack, architecture, constraints
3. `.ai/RULES.md` — mandatory engineering and safety rules
4. `.ai/MEMORY.md` — durable decisions, conventions, known pitfalls
5. `.ai/TASTE.md` — style and communication preferences
6. `.ai/MODEL_ROUTING.md` and `.ai/model-routing.yaml` — model tiers,
   task scoring, token budgets, and delegation protocol
7. `.ai/WORKFLOW.md` — task intake to human review flow

## Naming note

This root `AGENTS.md` follows the Codex convention: a short, repo-level
entry point that any AI coding agent reads automatically. ForgeAI's own
agent role definitions and model-routing strategy live separately in
`.ai/AGENT_REGISTRY.md` (with per-role detail in `.ai/agents/*.md`) — read
that for sub-agent responsibilities, not this file.

If `.ai/PROJECT.md` still contains `TODO` placeholders, follow
`.ai/BOOTSTRAP.md` before starting implementation work.