eryao/AGENTS.md

# Project AGENTS Router

Root `AGENTS.md` is routing + cross-domain policy only.
Do not place backend/frontend implementation details here.

## Scope

- Applies to repository root and cross-domain tasks.
- Subdomain rules: `backend/AGENTS.md`, `apps/AGENTS.md`.
- If rules conflict, use the stricter one.

## Rule Order

1. System / developer / platform safety instructions
2. Workspace runtime rules (`AGENTS.md` + `rules/*`)
3. This file (routing + project-level constraints)
4. Subdomain rules (backend/apps)

## Mandatory Routing

- `backend/**` must follow `backend/AGENTS.md`.
- `apps/**` must follow `apps/AGENTS.md`.
- Cross-domain changes must satisfy all relevant subdomain rules.
- `infra/**` follows this file plus `infra/` conventions.

## Project-Wide Constraints

- Default development branch is `dev`; do not develop directly on `main`.
- Never push unless explicitly requested by the user.
- Keep AGENTS layered and lean: shared rules at root, domain rules in sub-AGENTS.
- **No Error Swallowing**: All exceptions must propagate or be converted to typed errors. Never catch an exception, log it, and silently continue. This destroys debuggability.

## Git Safety (CRITICAL)

- **NEVER execute `git checkout -- <file>` or any git command that modifies files without explicit user approval.**
- **NEVER reset, revert, or discard uncommitted changes without user consent.**
- If you need to discard changes, ask the user first and explain exactly what will be lost.
- Before any destructive git operation, list the affected files and get confirmation.
- This rule is non-negotiable. Violation will cause irreversible loss of user work.

## Protocol Source of Truth

`docs/protocols/` is the single source of truth for protocol and data format.

- Update protocol docs before changing data/API/UI contracts.
- Document compatibility strategy (backward-compatible vs migration).
- Keep frontend/backend implementations aligned with documented protocol.

## Database Access

When viewing data in the database, use `supabase mcp` tools (`supabase_execute_sql`, `supabase_list_tables`, etc.) instead of direct queries or other methods.

## Image Handling

When reading images, check whether the model has native multimodal capability first. If it does, use `Read` tool to read images directly. If it does not, fall back to `understand_image` tool. Only use `Read` tool for non-image files.

<!-- TRELLIS:START -->
# Trellis Instructions

These instructions are for AI assistants working in this project.

This project is managed by Trellis. The working knowledge you need lives under `.trellis/`:

- `.trellis/workflow.md` — development phases, when to create tasks, skill routing
- `.trellis/spec/` — package- and layer-scoped coding guidelines (read before writing code in a given layer)
- `.trellis/workspace/` — per-developer journals and session traces
- `.trellis/tasks/` — active and archived tasks (PRDs, research, jsonl context)

If a Trellis command is available on your platform (e.g. `/trellis:finish-work`, `/trellis:continue`), prefer it over manual steps. Not every platform exposes every command.

If you're using Codex or another agent-capable tool, additional project-scoped helpers may live in:
- `.agents/skills/` — reusable Trellis skills
- `.codex/agents/` — optional custom subagents

## Subagents

- ALWAYS wait for every spawned subagent to reach a terminal status before yielding, acting on partial results, or spawning followups.
  - On Codex, this means calling the `wait` tool with the subagent's thread id (requires `multi_agent_v2`). Do NOT infer completion from elapsed time.
  - On Claude Code / OpenCode, this means awaiting the Task/agent tool result before continuing.
- NEVER cancel or re-spawn a subagent that hasn't finished. If a subagent appears stuck, raise the wait timeout (Codex default 30s, max 1h) before judging it broken.
- Spawn subagents automatically when:
  - Parallelizable work (e.g., install + verify, npm test + typecheck, multiple tasks from plan)
  - Long-running or blocking tasks where a worker can run independently
  - Isolation for risky changes or checks

Managed by Trellis. Edits outside this block are preserved; edits inside may be overwritten by a future `trellis update`.

<!-- TRELLIS:END -->