growi / rules

---
description: How Cursor Agent should use this repo's Claude Code knowledge (AGENTS.md, .claude, skills) without duplicating it
alwaysApply: true
---

# Single source of truth: `.claude/` + `AGENTS.md`

This repository is maintained for Claude Code. Cursor Agent must consume the same
files — never maintain a parallel copy of instructions, skills, or rules.

## Already native to Cursor

| Artifact | Behavior |
|---|---|
| `AGENTS.md` (root + nested) | Auto-loaded by Cursor |
| `CLAUDE.md` (points at `AGENTS.md` + Kiro workflow) | Treat as entrypoint / workflow overlay |
| `.claude/skills/**/SKILL.md` | Offered as Agent Skills — read and follow when relevant |
| `apps/app/.claude/skills/**/SKILL.md` | Same, when working under `apps/app/` |

## Bridged via `sessionStart` hook (not `@` includes)

Claude Code auto-injects `.claude/rules/`. Cursor does **not**.

Cursor docs claim `@filename` inside `.mdc` rules attaches file contents, but that
mechanism does not work yet. Copying rule bodies into `.cursor/rules/` would be
dual management — do not do that.

Instead, `.cursor/hooks.json` runs `sessionStart` →
`.cursor/hooks/inject-claude-rules.mjs`, which reads the canonical markdown under
`.claude/rules/` and `apps/app/.claude/rules/` and returns them as
`additional_context`. **Edit only the `.claude/` sources.**

## When to load more

- Skills: if there is even a small chance a listed skill applies, read its `SKILL.md` first
- Package docs: `apps/*/CLAUDE.md` / `AGENTS.md` when touching that package
- Specs / steering: `.kiro/specs/`, `.kiro/steering/` per `CLAUDE.md`
- Slash-style workflows in `.claude/commands/` and kiro skills: follow the matching
  skill under `.claude/skills/` rather than inventing a Cursor-only procedure

More agent context in growilabs/growi

27 other files this repository gives its agents.