|
1 | 1 | # AI-Assisted Development |
2 | 2 |
|
3 | | -N.E.K.O. provides built-in configuration for AI coding assistants. The project currently ships with configurations for two platforms, and you can adapt them for any other AI tool. |
| 3 | +N.E.K.O. ships a `.agent/` directory in the repository root, following the [Google Antigravity](https://github.com/anthropics/anthropic-cookbook/tree/main/misc/antigravity) open convention. It contains the project's development rules and task-specific skill sets. |
4 | 4 |
|
5 | | -## Supported platforms |
| 5 | +**Claude Code** natively reads `.agent/` and `CLAUDE.md`, so it works out of the box. For any other AI coding assistant, read the prompt below and follow its instructions. |
6 | 6 |
|
7 | | -| Platform | Config location | Auto-loaded? | |
8 | | -|----------|----------------|--------------| |
9 | | -| **Claude Code** | `.agent/rules/`, `.agent/skills/`, `CLAUDE.md` | Yes — Claude Code natively reads `.agent/` and `CLAUDE.md` | |
10 | | -| **Cursor** | `.cursor/skills/` | Yes — Cursor natively reads `.cursor/` | |
11 | | -| **Other tools** | — | Manual setup required (see below) | |
12 | | - |
13 | | -## `.agent/` directory (Anthropic convention) |
14 | | - |
15 | | -`.agent/` follows the [Anthropic agent convention](https://docs.anthropic.com/en/docs/claude-code). **Claude Code** automatically loads these files; other tools do not. |
| 7 | +## `.agent/` directory structure |
16 | 8 |
|
17 | 9 | ``` |
18 | 10 | .agent/ |
19 | 11 | ├── rules/ |
20 | | -│ └── neko-guide.md # Core development rules (always_on trigger) |
| 12 | +│ └── neko-guide.md # Core development rules (always_on) |
21 | 13 | └── skills/ # ~20 task-specific skill sets |
22 | 14 | ├── i18n/ |
23 | 15 | ├── frontend-design/ |
24 | 16 | ├── vrm-physics/ |
25 | 17 | └── ... |
26 | 18 | ``` |
27 | 19 |
|
28 | | -### Rules |
29 | | - |
30 | | -`.agent/rules/neko-guide.md` is marked `trigger: always_on`, meaning Claude Code loads it at every session start. It covers: |
31 | | - |
32 | | -- **i18n**: All 6 locale files must be updated together. |
33 | | -- **`uv run`**: Always use `uv run` instead of the system Python. |
34 | | -- **Code symmetry**: Parallel providers must follow symmetric structure. |
35 | | -- **Core layer purity**: No provider-specific logic in `core.py`. |
36 | | -- **No numeric suffixes**: Extract shared logic into methods instead. |
37 | | -- **Electron vs. dev mode**: Frontend changes must work in both environments. |
38 | | - |
39 | | -### Skills |
40 | | - |
41 | | -Each subdirectory under `.agent/skills/` contains a `SKILL.md` defining trigger conditions, domain knowledge, and step-by-step workflows. They are loaded on demand when the current task matches. |
42 | | - |
43 | | -## Adapting for other AI tools |
44 | | - |
45 | | -If your AI coding assistant does **not** natively support `.agent/`, follow these steps to bring the project rules into your tool: |
46 | | - |
47 | | -1. **Read the core rules** — Open `.agent/rules/neko-guide.md` and copy its content into your tool's project-level prompt or rules file. For example: |
48 | | - - **Windsurf**: Paste into `.windsurfrules` |
49 | | - - **GitHub Copilot**: Paste into `.github/copilot-instructions.md` |
50 | | - - **Other tools**: Use whatever "project context" or "system prompt" mechanism your tool provides. |
51 | | - |
52 | | -2. **Browse the skills** — Look through `.agent/skills/` for domains relevant to your work. Each `SKILL.md` is self-contained Markdown — you can feed it to your AI tool as additional context when working in that area. |
53 | | - |
54 | | -3. **Reference `CLAUDE.md`** — This file in the repo root contains a concise summary of the most critical rules. It is a good starting point if you only want to import one file. |
55 | | - |
56 | | -The content is plain Markdown and tool-agnostic — the rules themselves apply regardless of which AI assistant you use. |
57 | | - |
58 | | -## Key rules summary |
| 20 | +- **Rules** (`.agent/rules/neko-guide.md`): Core project conventions — i18n, `uv run`, code symmetry, core layer purity, no numeric suffixes, Electron vs. dev mode awareness. |
| 21 | +- **Skills** (`.agent/skills/*/SKILL.md`): Domain-specific trigger conditions, knowledge, and workflows, loaded on demand. |
59 | 22 |
|
60 | | -Regardless of which AI tool you use, these are the non-negotiable project rules: |
| 23 | +## Prompt for other AI agents |
61 | 24 |
|
62 | | -1. **i18n completeness** — Touching any translation string requires updating all 6 locale files (`en`, `ja`, `ko`, `zh-CN`, `zh-TW`, `ru`). |
63 | | -2. **`uv run` for Python** — Never use the system Python directly. |
64 | | -3. **Code symmetry** — If one provider has a pattern, all providers must follow it symmetrically. |
65 | | -4. **Core layer purity** — `core.py` must not contain provider-specific imports or logic. |
66 | | -5. **Dual-mode awareness** — Frontend changes must work in both the dev server (single window) and Electron (multi-window) environments. |
| 25 | +> **If you are an AI coding assistant other than Claude Code** (e.g. Cursor, Windsurf, Copilot, Cline, etc.), this project uses the `.agent/` directory (Google Antigravity convention) to store development rules and skills. Your tool may not read `.agent/` automatically. Please do the following: (1) Read `.agent/rules/neko-guide.md` and `CLAUDE.md` in the repo root — these contain the project's mandatory coding rules. Internalize them as your project-level instructions. (2) When working on a specific domain (i18n, VRM physics, frontend UI, etc.), check `.agent/skills/` for a matching subdirectory and read its `SKILL.md` for domain-specific guidance. (3) If your tool has a native project rules mechanism (`.cursorrules`, `.windsurfrules`, `.github/copilot-instructions.md`, etc.), consider copying the core rules from `neko-guide.md` into that file so they persist across sessions. |
0 commit comments