docs: rewrite CLAUDE.md with project architecture and commands

Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
This commit is contained in:
Naiyuan Qing 2026-01-30 21:50:12 +08:00
parent 1a1c3ec086
commit 456d9c5429

149
CLAUDE.md
View file

@ -1,63 +1,102 @@
# Project Instructions for AI Agents # CLAUDE.md
This file provides guidance to Claude Code (claude.ai/code) when working with code in this repository.
## Project Overview
Super Multica is a distributed AI agent framework with a monorepo architecture. It includes an agent engine with multi-provider LLM support, a WebSocket gateway, a console hub for multi-agent coordination, and frontend apps (Next.js web, Electron desktop).
## Monorepo Structure
- **`src/`** — Core modules (agent engine, gateway, console, client, shared types)
- **`apps/web`** — Next.js 16 web app (`@multica/web`, port 3001)
- **`apps/desktop`** — Electron + Vite + React desktop app (`@multica/desktop`)
- **`packages/ui`** — Shared UI component library (`@multica/ui`, Shadcn/Tailwind CSS v4)
- **`packages/sdk`** — Gateway client SDK (`@multica/sdk`, Socket.io)
- **`packages/store`** — Zustand state management (`@multica/store`)
- **`skills/`** — Bundled agent skills (commit, code-review, skill-creator)
## Common Commands
```bash
# Install dependencies
pnpm install
# Development (all services concurrently: gateway:3000, console, web:3001)
pnpm dev
# Individual services
pnpm dev:gateway # WebSocket gateway only
pnpm dev:console # NestJS console with agent
pnpm dev:web # Next.js web app
pnpm dev:desktop # Electron desktop app
# Agent CLI
pnpm agent:cli # Non-interactive agent
pnpm agent:interactive # Interactive REPL mode
# Build (turbo-orchestrated)
pnpm build
# Type checking
pnpm typecheck
# Testing (vitest, tests live in src/**/*.test.ts)
pnpm test # Single run
pnpm test:watch # Watch mode
pnpm test:coverage # With v8 coverage
```
## Architecture
```
Frontend (web:3001 / desktop)
@multica/sdk (GatewayClient, Socket.io)
→ Gateway (NestJS, WebSocket, port 3000)
→ Console Hub (multi-agent coordination)
→ Agent Engine (LLM runner, sessions, skills, tools)
```
**Agent Engine** (`src/agent/`): Orchestrates LLM interactions with multi-provider support (OpenAI, Anthropic, DeepSeek, Kimi, Groq, Mistral, Google, Together). Features session management (JSONL-based, UUIDv7 IDs), profile system (`~/.super-multica/agent-profiles/`), modular skills with hot-reload, and token-aware context window guards (compaction modes: tokens, count, summary).
**Gateway** (`src/gateway/`): NestJS WebSocket server with Socket.io for real-time message passing, RPC request/response, and streaming.
**Console** (`src/console/`): NestJS hub for multi-agent coordination with a web dashboard.
## Tech Stack & Config
- **Package manager**: pnpm 10 with workspaces (`pnpm-workspace.yaml`)
- **Build orchestration**: Turborepo (`turbo.json`)
- **TypeScript**: ESNext target, NodeNext modules, strict mode, `verbatimModuleSyntax`, `experimentalDecorators` (NestJS)
- **Testing**: Vitest with globals enabled, node environment
- **Frontend**: React 19, Next.js 16, Tailwind CSS v4, Shadcn/UI (zinc base, hugeicons)
- **Backend**: NestJS 11, Socket.io, Pino logging
- **CLI bundling**: esbuild → `bin/` directory
## Environment Setup
```bash
cp .env.example .env
# Set LLM_PROVIDER and corresponding API keys
# Supported: openai, anthropic, deepseek, kimi-coding, groq, mistral, together, google
# .env is loaded automatically via --env-file in all dev/agent scripts
```
## Atomic Commits ## Atomic Commits
After completing any task that modifies code, you MUST create atomic commits before ending the conversation. Do not ask for permission - just do it. After completing any task that modifies code, create atomic commits without asking for permission.
### Workflow 1. Run `git status` and `git diff` to see all modifications
2. Skip if no changes exist
3. Group changes by logical purpose (feature, fix, refactor, docs, test, chore)
4. Stage and commit each group separately
1. **Check for changes**: Run `git status` and `git diff` to see all modifications **Format**: Conventional commits — `<type>(<scope>): <description>`
2. **Skip if clean**: If there are no changes, skip the commit process
3. **Analyze changes**: Group changes by their logical purpose:
- Feature additions
- Bug fixes
- Refactoring
- Documentation
- Tests
- Configuration/dependencies
4. **Create atomic commits**: For each logical group, stage only the relevant files and create a separate commit
### Commit Process Types: `feat`, `fix`, `refactor`, `docs`, `test`, `chore`
For each logical group of changes:
```bash
# Stage specific files for this logical change
git add <file1> <file2>
# Commit with conventional commit message
git commit -m "<type>(<scope>): <description>"
```
### Commit Message Format
Use conventional commits:
- `feat`: New feature
- `fix`: Bug fix
- `refactor`: Code refactoring (no functional change)
- `docs`: Documentation changes
- `test`: Adding or updating tests
- `chore`: Build, config, dependencies
### Examples
If you modified:
- `src/api/user.ts` (added new endpoint)
- `src/api/user.test.ts` (tests for new endpoint)
- `src/utils/format.ts` (refactored helper)
- `README.md` (updated docs)
Create three commits:
1. `git add src/api/user.ts src/api/user.test.ts && git commit -m "feat(api): add user profile endpoint"`
2. `git add src/utils/format.ts && git commit -m "refactor(utils): simplify date formatting logic"`
3. `git add README.md && git commit -m "docs: update API documentation"`
### Rules
**Rules**:
- Each commit should be independently meaningful and buildable - Each commit should be independently meaningful and buildable
- Related test files should be committed with their implementation - Related test files go with their implementation
- Never create empty commits - Never create empty commits or combine unrelated changes
- Never combine unrelated changes in one commit - `git commit --amend` only for immediate small fixes to the last commit
- Keep commit messages concise but descriptive
- If all changes are related to one logical unit, a single commit is fine
- `git commit --amend` can be used for immediate small fixes to the last commit, but not for unrelated changes