software-engineering-skills
v1.0.0
Published
Multi-agent engineering pipeline: REFINE, PLAN, BUILD, TEST, REVIEW. Orchestrates end-to-end feature development with TDD, incremental builds, and code health analysis. Language-agnostic.
Downloads
132
Maintainers
Readme
Engineering Skills Plugin
Multi-agent engineering pipeline for end-to-end feature development.
Pipeline
REFINE → PLAN → BUILD → TEST → REVIEWEach stage is a dedicated agent. A single orchestrator command chains them sequentially with human approval gates between each stage.
Usage
Run the full pipeline:
/swe:pipeline <feature description>Stages
| Stage | Agent | Purpose |
|-------|-------|---------|
| REFINE | refiner | Clarify the idea, resolve ambiguities, produce feature definition |
| PLAN | planner | Break into atomic tasks, produce reviewable markdown plan |
| BUILD | builder | TDD implementation — one task at a time (red-green-refactor) |
| TEST | tester | Evolving test plan with iterative execution and status tracking |
| REVIEW | reviewer | Code review + silent failure hunting + code health analysis |
Design Principles
- Language-agnostic: Adapts to whatever language, framework, and build tool the project uses
- Context-aware: Reads CLAUDE.md and
.claude/config to discover project conventions - Human gates: Every stage requires explicit approval before proceeding
- Test-driven: BUILD stage follows red-green-refactor cycle
- Living documents: PLAN produces a task plan doc, TEST produces an evolving test plan doc
- Read-only review: REVIEW agent never modifies code — produces findings only
Agents
Each agent can also be invoked independently outside the full pipeline.
refiner (opus, blue)
Explores the codebase, asks clarifying questions, produces a structured feature definition.
planner (opus, green)
Decomposes into atomic tasks with dependencies, writes a markdown plan document for review.
builder (opus, yellow)
Implements tasks one at a time using TDD. Reads project conventions and adapts to any build tool.
tester (sonnet, cyan)
Creates and maintains a living test plan document. Runs iterative test rounds and tracks pass/fail status.
reviewer (opus, red)
Combines three review dimensions: code review (confidence-scored), silent failure hunting, and code health analysis. Never modifies code.
