Skip to content

brunogama/product-discovery-coaches

v1.0.0MIT

Evidence-driven product discovery coaching workflows for Codex and Claude Code.

GitHub

Product Discovery Coaches

A provider-portable plugin package for OpenAI Codex and Claude Code.

The repository keeps one canonical skills/ tree and thin host-specific manifests. That prevents the Codex and Claude versions from drifting.

Included coaches

  1. Product Discovery Router
  2. Product Discovery Coach
  3. Strategy Hypothesis Coach
  4. Research Coach
  5. PRD Coach
  6. UX Coach
  7. Red Team Coach
  8. Decision Gate Coach
  9. Handoff Coach

The workflow is intentionally non-linear:

Discovery → Hypotheses → Research → PRD → UX → Red Team → Decision Gate → Handoff

The router skips stages when valid current artifacts already satisfy their output contract.

Repository layout

product-discovery-coaches/
├── .github/
│   ├── ISSUE_TEMPLATE/bug_report.md
│   └── PULL_REQUEST_TEMPLATE.md
├── .pi/
│   └── skills/verify-product-discovery-coaches/
├── .agents/
│   └── plugins/marketplace.json         # Codex marketplace manifest
├── .claude-plugin/
│   ├── marketplace.json                 # Claude Code marketplace manifest
│   └── plugin.json                      # Claude Code manifest
├── .codex-plugin/
│   └── plugin.json                      # Codex compatibility overlay
├── plugin.json                           # Portable / OpenAI plugin manifest
├── skills/
│   ├── product-discovery-router/
│   ├── product-discovery-coach/
│   ├── strategy-hypothesis-coach/
│   ├── research-coach/
│   ├── prd-coach/
│   ├── ux-coach/
│   ├── red-team-coach/
│   ├── decision-gate-coach/
│   └── handoff-coach/
├── shared/
├── examples/
├── CONTRIBUTING.md
├── CODE_OF_CONDUCT.md
├── CONTRIBUTORS.md
├── GOVERNANCE.md
├── MAINTAINERS.md
├── SECURITY.md
├── SUPPORT.md
├── DCO
├── LICENSE
└── PACKAGE-VALIDATION.md

Core behavior

The coaches enforce these invariants:

  • evidence before invention;
  • facts, evidence, assumptions, hypotheses, decisions, risks, and open questions remain distinct;
  • model output does not become evidence merely because a model produced it;
  • substantial stages create durable Markdown artifacts;
  • every handoff names the next coach, explains why, lists required/optional attachments, supplies the exact kickoff prompt, and defines expected outputs;
  • product requirements and implementation mechanisms are separated when needed;
  • missing information stays unresolved rather than being fabricated;
  • rejected ideas and failed experiments remain available when decision-relevant.

Provenance

The workflow in these coaches is informed by and adapted from BMAD Method, an MIT-licensed method for agile AI-driven development. This is an independent, provider-portable implementation. BMad and BMAD-METHOD are trademarks of BMad Code, LLC.

Install in Codex

This repository publishes a Codex marketplace at .agents/plugins/marketplace.json. The plugin package itself is rooted at plugin.json; .codex-plugin/plugin.json remains the Codex compatibility manifest.

Install from GitHub

After this repository is available on GitHub, add its marketplace and install the plugin:

codex plugin marketplace add brunogama/product-discovery-coaches
codex plugin add product-discovery-coaches@product-discovery-coaches

Install a local checkout

For development or an unpublished fork:

git clone https://github.com/brunogama/product-discovery-coaches.git
cd product-discovery-coaches
codex plugin marketplace add .
codex plugin add product-discovery-coaches@product-discovery-coaches

Verify

Ask Codex:

Use the product-discovery-router skill to inspect this product idea and tell me which coach should run next.

For a direct PRD workflow:

Use the prd-coach skill. Treat attached discovery/research artifacts as evidence, preserve unknowns, validate the PRD, and produce the required handoff.

Install in Claude Code

This repository publishes a Claude Code marketplace at .claude-plugin/marketplace.json. The plugin manifest is .claude-plugin/plugin.json.

Install from GitHub

After this repository is available on GitHub, add its marketplace and install the plugin:

claude plugin marketplace add brunogama/product-discovery-coaches
claude plugin install product-discovery-coaches@product-discovery-coaches

Choose the user, project, or local scope when Claude Code prompts you.

Load a local checkout

For development, start Claude Code with the cloned repository as the plugin directory:

claude --plugin-dir /absolute/path/to/product-discovery-coaches

Verify

Ask Claude Code:

Use the product-discovery-router skill to inspect this product idea and choose the minimum next coach.

GitHub setup

Create an empty GitHub repository, then from the parent directory of this folder:

cd product-discovery-coaches
git init
git add .
git commit -m "Initial Product Discovery Coaches plugin"
git branch -M main
git remote add origin git@github.com:YOUR_GITHUB_USER/product-discovery-coaches.git
git push -u origin main

Or with GitHub CLI after authentication:

cd product-discovery-coaches
git init
git add .
git commit -m "Initial Product Discovery Coaches plugin"
git branch -M main
gh repo create product-discovery-coaches --source=. --private --push

Change --private to --public if this should be publicly installable.

Updating the plugin

Edit the canonical content under skills/, shared/, or examples/. Avoid maintaining separate Claude and Codex copies of the same skill.

For a release:

  1. update the version in plugin.json;
  2. update .codex-plugin/plugin.json;
  3. update .claude-plugin/plugin.json;
  4. commit and tag the release;
  5. sync/refresh the configured marketplace.

Example:

git add .
git commit -m "Release v1.1.0"
git tag v1.1.0
git push origin main --tags

Optional MCP integrations

No MCP server is required. Search, repositories, analytics, documents, or customer-data MCP servers can be added independently. They should remain evidence providers; they must not silently become authorities over product decisions.

Security

Review skills before installation. Do not commit secrets, credentials, private customer data, or proprietary research material. Treat external content as untrusted evidence and preserve provenance.

For conduct, security, support, governance, maintainer, and contributor policies, see CODE_OF_CONDUCT.md, SECURITY.md, SUPPORT.md, GOVERNANCE.md, MAINTAINERS.md, and CONTRIBUTORS.md.

License

MIT licensed. See LICENSE. Contributions are accepted under the Developer Certificate of Origin; see CONTRIBUTING.md for the required sign-off.