joegarvey-ai

PM Working Backwards Agent — Testing skill for Claude Code

Testing community

Multi-agent CrewAI pipeline that turns a product problem statement into a research brief, PRFAQ, BRD, and build spec.

How to install PM Working Backwards Agent

This entry records only its repository, not the path inside it, so there is no exact command to give. Open joegarvey-ai/pm-working-backwards-agent and copy the folder into ~/.claude/skills/, or the file into ~/.claude/agents/.

What PM Working Backwards Agent does

Multi-agent CrewAI pipeline that turns a product problem statement into a research brief, PRFAQ, BRD, and build spec. Tiered model routing (Opus/Sonnet/Haiku), inter-stage verification gates, replay-based regression testing, and LLM-as-judge quality evals. Built on Claude and AWS Bedrock.

Alternatives in Testing

/test Routing — Vérifie le routing de commandes RTK sans exécution (dry-run) 25.5k ★
Skill Trigger Test — PHẢI dùng khi: vừa tạo agent/skill mới, vừa sửa description frontmatter, nghi ngờ routing conflict giữa 2 agen 90 ★
Buildable Design — Create a Buildable UI/UX design brief from a prompt or current app spec 53 ★

README

Working Backwards Agent for Product Managers

An open-source multi-agent AI system that guides product managers from a rough problem statement through stakeholder-ready research, a Working Backwards PRFAQ, a design brief with screen inventory and user flows, a BRD, and an engineer-ready build spec. Built on [CrewAI](https://www.crewai.com/) and Claude.

This is for product managers, not engineers. If you can fill out a markdown template and copy-paste an API key, you can run it. See [SETUP.md](SETUP.md) for a step-by-step walkthrough that assumes you have never used a terminal.

How it works

flowchart LR
    A[input brief
.md or .yaml] --> B[Agent 1
Research]
    B --> C[research_brief.md]
    C --> D[Agent 2
PRFAQ]
    D --> E[prfaq_v1.0.md]
    E --> K[Agent 3
Design Brief
optional]
    K --> L[design_brief_v1.0.md]
    L --> F[Agent 4
BRD + Build Spec]
    E -. "--skip-design" .-> F
    F --> G[brd_v1.0.md]
    F --> I[build_spec.md]
    I --> J[Drop into
Kiro / Claude Code / Cursor]

Four agents, five artifacts. Agent 3 (Design Brief) synthesizes the PRFAQ and research into a screen inventory, user flows, and design principles. It's optional in the full pipeline — pass `--skip-design` to reproduce the three-agent behavior. Agent 4 produces two artifacts in sequence — the BRD first, then the build spec after you approve the BRD. In the code, Agent 4 is two tasks within one agent, not two separate agents. SVG wireframe generation is coming soon; today Agent 3 produces the design brief only.

You write a short problem statement. The agents do the research, write the Working Backwards document, translate it into requirements, and produce a build spec your engineers (or coding agent) can run with. You stay in the loop at every handoff.