SDLC orchestration layer that drives a feature from Jira ticket to production-ready Go code, tests, and documentation — using Claude as the execution engine for every stage.
You provide a Jira ticket (or a plain feature description). FlowPilot orchestrates five SDLC stages — each powered by Claude — and delivers a complete set of artifacts: architecture design documents, production Go code, code review findings with an auto-fix loop, Ginkgo v2 test files, and PR-ready documentation. Each stage validates its outputs before passing state to the next, so failures surface immediately rather than propagating silently through the pipeline.
┌─────────────────────────────────────────────────────────────────────┐
│ Feature SDLC Pipeline │
├──────────┬──────────┬─────────────┬──────────┬──────────────────────┤
│ Phase 1 │ Phase 2 │ Phase 2.5 │ Phase 3 │ Phase 4 │
│ DESIGN │ DEVELOP │ CODE REVIEW │ TESTING │ DOCUMENTATION │
├──────────┼──────────┼─────────────┼──────────┼──────────────────────┤
│ Arch │ Go code │ Blocking / │ Unit │ PR Summary │
│ analysis │ PR desc │ Warning │ Integr. │ Release Notes │
│ Risks │ API types│ findings │ E2E │ SHIP / JTBD docs │
│ Impl plan│ Tests │ Auto-fix ↺ │ Ginkgo v2│ │
└──────────┴──────────┴─────────────┴──────────┴──────────────────────┘
Input: Jira Ticket or Feature Description
Output: /output-dir with code/, tests/, design/, docs/
The code review stage includes an auto-fix loop. Blocking findings route the pipeline back to the development stage for a retry, up to MAX_REVIEW_ITERATIONS:
Development → Code Review ──PASS──→ Testing
│
FAIL (blocking findings)
│
└──→ Development (retry, max MAX_REVIEW_ITERATIONS)
| Phase | Stage Runner | Key Inputs | Key Outputs |
|---|---|---|---|
| 1 · Design | Design Agent | Jira ticket, repo paths | Architecture analysis, impacted components, risks, acceptance criteria, implementation plan |
| 2 · Development | Development Agent | Design analysis, implementation plan | Go source files, PR description, API types |
| 2.5 · Code Review | Code Review Agent | Generated code | BLOCKING/WARNING findings, pass/fail verdict, auto-fix loop |
| 3 · Testing | Testing Agent | Design + code | Unit, integration, e2e Ginkgo v2 test files, test plan |
| 4 · Documentation | Documentation Agent | All phase outputs | PR summary, release notes, SHIP docs, JTBD docs |
| Publish | publish.py | Output directory | GitHub PR, Jira comments/attachments |
Configure repository paths so stage runners can analyze actual Go types, CRDs, and controllers from source. Use repos.yaml for multi-repo setups:
cp repos.yaml.example repos.yaml # edit with your local clone pathsSee Configuration for details. Environment variables (SHIPWRIGHT_REPO_PATH, OPENSHIFT_BUILDS_REPO_PATH) are also supported for single-repo setups.
# 1. Install
git clone <repo> && cd muilti-agents-ocp-builds
uv venv && uv pip install -r requirements.txt
cp .env.example .env # fill in credentials
# 2. Run the dashboard
uv run python scripts/run_dashboard.py # http://localhost:8080Open http://localhost:8080 and click New Run. Enter a Jira ticket ID (e.g. BUILD-1707) or a feature description, then click Run Feature. The pipeline orchestrates all five stages and shows progress in real time.
For CLI and automation usage, see the CLI Reference.
See Quick Start for full credential setup (Vertex AI, Jira, GitHub).
output/BUILD-1707/
├── design/
│ ├── design_analysis.md # Architecture decisions, impacted components
│ └── implementation_plan.md # Step-by-step implementation guide
├── code/ # Production Go source files
├── tests/
│ ├── unit/ # Ginkgo v2 unit tests
│ ├── integration/ # Integration test files
│ └── e2e/ # End-to-end test files
├── docs/
│ ├── pr_description.md # Ready-to-use GitHub PR description
│ └── release_notes.md # Release notes for the feature
└── state.json # Full pipeline state snapshot
User Guide — full reference
| Section | Content |
|---|---|
| Quick Start | Dashboard-first setup: install, configure credentials, run |
| Dashboard | Web UI pages, real-time monitoring, session management |
| Architecture | Pipeline design, state management, security layers |
| Stage Runners | Per-stage reference |
| CLI Reference | All CLI commands for scripting and automation |
| API Reference | REST API endpoints, heartbeat protocol, database schema |
| Authentication | Vertex AI setup |
| Publishing | Push to GitHub / Jira |
| Requirement | Detail |
|---|---|
| Python | 3.12+ |
| Claude | Google Vertex AI (Application Default Credentials) |
| Jira | API token for ticket fetching |
| GitHub | PAT for PR enrichment and publishing |
| Qodo | Optional — Claude is used as fallback if not installed |