Milestones
List view
EventBus↔UI integration, stdio/SSE mode testing, E2E verification, Nerd Font fallback, performance testing, and documentation. Starts after Phase 3 completion.
No due date•5/5 issues closedImplement Header, AgentCard, ProgressBar, AgentTree, AgentGrid, StatusBar components and React Hooks state management. Starts after Phase 2 completion.
No due date•9/9 issues closed이벤트 타입 정의, TuiInterceptor 구현, EventBus 서비스 구현, EventBus↔React hooks 연결. Phase 1 완료 후 시작.
No due date•4/4 issues closedAdd Ink dependencies, create TUI directory structure, set up EventBus (EventEmitter2) module, parse --tui CLI flag and integrate with main.ts
No due date•5/5 issues closedCodingbuddy TUI Agent Monitor - AI 에이전트 활동을 실시간으로 시각화하는 터미널 대시보드. 설계 문서: docs/plans/2026-02-12-tui-agent-monitor-design.md
No due date# Milestone: Claude Code Plugin ## Overview A native Claude Code plugin that integrates codingbuddy rules, agents, and workflow modes directly into Claude Code's plugin ecosystem. **Location**: `apps/claude-code-plugin/` --- ## Why ### Current State Codingbuddy currently supports Claude Code through: 1. **MCP Server**: Provides rules via `codingbuddy mcp` command 2. **Adapter files**: `.claude/` directory with static CLAUDE.md instructions ### Limitations of Current Approach #### 1. MCP Server Overhead MCP requires: - Separate process management - Configuration in `~/.claude/settings.json` - Network communication overhead (stdio/SSE) For Claude Code specifically, a **native plugin** would be more direct and efficient. #### 2. Static Adapter Limitations `.claude/rules/` files are: - Read once at session start - Cannot dynamically respond to user input - No runtime context awareness #### 3. Missing Native Integration Current setup cannot: - Intercept user prompts for keyword detection (PLAN/ACT/EVAL) - Dynamically inject agent prompts based on context - Coordinate with other Claude Code plugins - Provide skills that users can invoke ### Why a Native Plugin? | Aspect | MCP Server | Native Plugin | |--------|-----------|---------------| | Startup | External process | Built-in | | Communication | stdio/SSE | Direct API | | Prompt interception | Not possible | Supported | | Skill registration | Not possible | Supported | | Plugin coordination | Limited | Full support | --- ## Goals ### Primary Goals 1. **Rules Injection**: Dynamically provide codingbuddy-rules to Claude Code context 2. **Agent Activation**: Generate and inject specialist agent prompts on demand 3. **Keyword Processing**: Parse PLAN/ACT/EVAL keywords and manage mode transitions 4. **Plugin Harmony**: Coordinate with other Claude Code plugins (e.g., superpowers) ### Secondary Goals 1. **Skill Registration**: Register invocable skills (e.g., `/plan`, `/eval`, `/activate-agent`) 2. **Context Awareness**: Adapt rules based on project detection 3. **Configuration Sync**: Share config with MCP server (`codingbuddy.config.json`) --- ## Core Features ### 1. Rules Provider ``` User prompt → Plugin intercepts → Injects relevant rules → Claude processes ``` - Dynamically inject rules from `packages/rules/.ai-rules/` - Context-aware rule selection based on current task - Respect `codingbuddy.config.json` settings ### 2. Agent Activation ``` User: "Activate Security Specialist" Plugin: Loads security-specialist.json → Generates activation prompt → Injects ``` - Load agent definitions from `.ai-rules/agents/` - Generate activation prompts with proper context - Support mode-specific agent behavior (planning/implementation/evaluation) ### 3. Keyword Processing ``` User: "PLAN authentication feature" Plugin: Detects PLAN keyword → Sets mode → Injects PLAN mode rules ``` - Intercept prompts starting with PLAN/ACT/EVAL - Manage mode state across conversation - Inject mode-specific instructions and rules ### 4. Plugin Coordination - Expose hooks for other plugins to access codingbuddy rules - Respect other plugins' skills and avoid conflicts - Provide composable primitives for plugin authors ### 5. Registered Skills | Skill | Description | |-------|-------------| | `/plan` | Enter PLAN mode with optional task description | | `/act` | Enter ACT mode to execute planned changes | | `/eval` | Enter EVAL mode for code review | | `/agent <name>` | Activate a specialist agent | | `/rules` | Show active rules and configuration | --- ## Architecture ``` apps/claude-code-plugin/ ├── src/ │ ├── index.ts # Plugin entry point │ ├── providers/ │ │ ├── rules.provider.ts # Rules injection logic │ │ └── agent.provider.ts # Agent activation logic │ ├── interceptors/ │ │ └── keyword.interceptor.ts # PLAN/ACT/EVAL detection │ ├── skills/ │ │ ├── plan.skill.ts │ │ ├── eval.skill.ts │ │ └── agent.skill.ts │ └── coordination/ │ └── plugin-bridge.ts # Inter-plugin communication ├── package.json └── tsconfig.json ``` --- ## Success Criteria 1. **Rules available in Claude Code** without MCP server running 2. **Keyword detection works** for PLAN/ACT/EVAL 3. **Agent activation** generates correct prompts 4. **Skills registered** and invocable via `/plan`, `/eval`, etc. 5. **No conflicts** with existing plugins (superpowers, etc.) --- ## Dependencies - `packages/rules/` (codingbuddy-rules) - Claude Code Plugin SDK (TBD based on official docs) - Shared config with `apps/mcp-server/` --- ## Open Questions 1. What is the official Claude Code Plugin SDK/API? 2. How to handle plugin conflicts and priority? 3. Should this replace or complement the MCP server approach? 4. How to sync state between MCP server and plugin? --- ## Related Documents - [MCP Server Architecture](../api.md) - [Keyword Invocation Design](../plans/2025-12-15-keyword-invocation-design.md) - [Agents README](../../packages/rules/.ai-rules/agents/README.md)
Due by January 31, 2026•1/1 issues closed