SDLC — Multi-Agent Development Pipeline
A Claude Code plugin that maps agile team roles to LLM agents, covering the full software development lifecycle.
Quick Start
cd /path/to/your-project
claude --plugin-dir /path/to/sdlcThen inside Claude Code:
/sdlc:develop Add user authentication with JWT tokens
Commands
| Command | Description |
|---|---|
/sdlc:develop <task> |
Full pipeline: plan → localize → implement → review |
/sdlc:plan <task> |
Break a task into sub-tasks with acceptance criteria |
/sdlc:review |
6-point code review on uncommitted changes |
/sdlc:summarize |
Generate natural-language summaries of the codebase |
How It Works
/sdlc:develop launches a Supervisor Agent that orchestrates five specialized agents:
User Task
│
▼
┌─────────────────┐
│ Summary Agent │ Scans codebase, generates NL summaries
└────────┬────────┘
▼
┌─────────────────┐
│ Sprint Agent │ Breaks task into sub-tasks + acceptance criteria
└────────┬────────┘
▼
┌─────────────────────────┐
│ For each sub-task: │
│ │
│ ┌─────────────────┐ │
│ │ Control Agent │ │ Localizes relevant code (Meta-RAG)
│ └────────┬────────┘ │
│ ▼ │
│ ┌─────────────────┐ │
│ │ Developer Agent │ │ Implements changes + writes tests
│ └────────┬────────┘ │
│ │ │
│ (retry up to 3x │
│ on test failure) │
└─────────────────────────┘
▼
┌─────────────────┐
│ Peer Agent │ 6-point code review
└─────────────────┘
| Agent | Role | Model |
|---|---|---|
| Supervisor | Pipeline orchestrator | opus |
| Summary | Codebase summarizer | sonnet |
| Sprint | Task decomposition + planning | sonnet |
| Control | Code localization (Meta-RAG) | sonnet |
| Developer | Implementation + testing | opus |
| Peer | Code review (6-point checklist) | opus |
Peer Review Checklist
The Peer Agent evaluates changes against six criteria:
- Functionality Alignment — Do changes meet acceptance criteria?
- Vulnerability Check — Any security issues?
- Performance — N+1 queries, memory leaks, etc.?
- Hallucination Detection — References to non-existent APIs or methods?
- Code Quality — Naming, structure, DRY, SOLID
- Test Coverage — Are tests sufficient?
Codebase Summaries
The Summary Agent generates hierarchical natural-language summaries stored in .sdlc/summaries/ inside your project. These summaries enable the Control Agent to quickly localize relevant code without reading every file.
.sdlc/summaries/
├── src/
│ ├── services/
│ │ └── auth.ts.summary.md
│ └── routes/
│ └── auth.routes.ts.summary.md
└── ...
Summaries are created on first run and incrementally updated on subsequent runs. Add .sdlc/ to your .gitignore if you don't want to track them.
File Structure
sdlc/
├── .claude-plugin/
│ └── plugin.json
├── agents/
│ ├── supervisor.md # Pipeline orchestrator
│ ├── summary.md # Codebase summarizer
│ ├── sprint.md # Task decomposition
│ ├── control.md # Code localizer (Meta-RAG)
│ ├── developer.md # Code implementation
│ └── peer.md # Code reviewer
├── skills/
│ ├── develop/SKILL.md # /develop — full pipeline
│ ├── plan/SKILL.md # /plan — planning only
│ ├── review/SKILL.md # /review — review only
│ └── summarize/SKILL.md # /summarize — summaries only
└── scripts/
└── collect-summaries.sh
Error Handling
- If the Sprint Agent finds the task ambiguous, it will ask clarification questions before proceeding.
- If the Developer Agent's tests fail, the Supervisor retries up to 3 times with re-localization.
- If the Peer Agent verdict is "NEEDS CHANGES", action items are reported for you to decide on.
- The pipeline never commits to git — you stay in control.
License
MIT