One factory. Four products. A pipeline you can govern, verify and watch.
The Factory family turns an idea into shipped, tested software through a chain of autonomous services — each useful on its own, each designed to hand off to the next, all following the PARR loop: Prepare · Act · Reflect · Review.
PFactory ──▶ AIFactory ──▶ TFactory
(Plan) (Act) (Verify)
└───────────── observed & steered by ─────────────┘
CFactory
This is the repository for the whole program — cross-cutting plans, the shared pipeline, and the place the four products come together.
The products
Prepare / Plan + ReviewPFactory — governed planning, grounded in your infrastructure
The planning layer that sits in front of coding agents. It ingests plans, enriches them with live organizational context (Kubernetes, AWS/Azure/GCP, Backstage, internal wikis), runs architecture / security / feasibility review gates with citations, records human approval, and emits governed GitHub issues.
- Context-grounded planning from real cloud + catalog state
- Hybrid deterministic + LLM review gates, every verdict cited
- Human-approval gate before any work is emitted
- Kanban board, feasibility & cost estimates, living templates
ActAIFactory — spec-first execution that ships merge-ready code
The execution engine. A planner writes a reviewable spec, a coder implements it in an isolated git worktree, and a QA agent validates against the spec — multi-provider, able to delegate to GitHub Copilot or GitLab Duo, and enterprise-grade.
- Spec-first: every run starts from a written, editable spec
- Isolated worktrees — nothing touches main until you merge
- Multi-provider, per-phase model selection; MCP control plane
- Enterprise: SAML/SCIM, tenant isolation, audit, LiteLLM gateway
Reflect / ReviewTFactory — tests you can trust, not just a green bar
Autonomous test generation + execution across modality lanes (unit, browser, API, integration, mutation). It grades every generated test on a 5-signal verdict — coverage delta, stability re-runs, mutation kills, lint, semantic relevance — and posts a ranked triage report to your PR.
- Five-signal verdict: meaningful tests, not coverage theatre
- Modality lanes with real evidence (screenshots, video, HAR, mutants)
- Bidirectional handback: failures route back to AIFactory’s QA fixer
- Works from any AC source (markdown / Gherkin / EARS) or MCP
Review / Observe & Steer — newCFactory — the control tower over all three
The newest member and the piece that turns the others into a suite. CFactory
threads every unit of work across the three services into one WorkItem (keyed by
GitHub issue), shows it on a single live cockpit, and adds an agentic copilot
that explains pipeline state and proposes human-confirmed actions.
- One pane of glass: where is every feature across plan → code → test
- Agentic copilot: “why is #182 stuck?”, answered from real cross-service state
- Advise + confirm: the copilot prepares actions; a human always clicks
- Built on the family skeleton; reuses AIFactory’s enterprise security
CFactory is early — screenshots land as the cockpit ships. Follow the repo and its roadmap.
How they cooperate
The products are independently useful, but their real power is the handoff chain — the cooperation we’re building out:
- PFactory → AIFactory. PFactory emits governed GitHub issues; AIFactory picks them up and builds, carrying the issue number as provenance.
- AIFactory → TFactory. A finished feature on a branch is handed to TFactory, which generates and grades a test suite against the acceptance criteria.
- TFactory → AIFactory (handback). When tests fail, TFactory routes a correction request back to AIFactory’s QA fixer — a bounded, closed loop.
- CFactory over everything. A shared correlation key — the GitHub issue
number — threads
plan → code → branch/PR → tests, so CFactory can show and steer the whole pipeline from one place.
PFactory ──issues──▶ AIFactory ──branch/PR──▶ TFactory
▲ │ ▲ │
│ │ └──── handback ───────┘
└─ correlation key (GitHub issue #) ──────────┘
every step observed by CFactory
The connective tissue — a shared correlation key, a normalized completion-event schema, and a canonical local port map — is tracked as the PARR-spine epic in this repo.