product-eval

product-eval

Find what's worth building next, with the evidence to prove it.

product-eval discovers and analyzes real customer problems from the tools where they already live: support tickets (Zendesk), CRM (Salesforce, HubSpot), issue trackers (Jira), product analytics, and reviews. It weighs the evidence, ranks what matters, and builds a prioritized, evidence-backed narrative of the problem worth solving.

When a bet is ready, hand it off: export a brief or decision memo, or pass a build spec straight to a wireframing tool or an AI coding agent like Codex or Claude Code to ship an MVP. It runs in both Claude Code and Codex, and keeps the rigor (evidence weighting, readiness gates, audit files) behind a simple, outcome-first chat.

It helps you answer four practical questions:

• Rank opportunities: what should we build next?
• Decide on one bet: is this worth building now, or do we need more proof?
• Pressure-test a plan: where is this roadmap, PRD, or strategy weak before we commit?
• Write the output: turn an approved decision into a brief, decision memo, build spec, or visual scorecard.

The plugin keeps the rigor behind the scenes: it can discover available sources, gather and weight evidence, synthesize themes into buildable problems, score Value and Confidence, check decision readiness, and produce decision artifacts. Users should not need to choose those internal steps directly; they can describe the outcome they want.

Install

Cowork: open the .plugin file. It renders as an installable card; click to add it.

Claude Code: add this directory or its hosting repo as a plugin marketplace, then install:

/plugin marketplace add <path-or-repo>
/plugin install product-eval

Codex: this source is Codex-compatible through .codex-plugin/plugin.json. For Codex marketplace installation, place or symlink this folder as <marketplace-root>/plugins/product-eval, make sure that marketplace root is registered with Codex, then install product-eval from that marketplace. This repo cleanup does not create or edit global Codex marketplace config such as ~/.agents/plugins/marketplace.json.

Quick install (either tool, no marketplace): clone this repo and run ./install.sh. It symlinks the 14 skills into ~/.claude/skills and ~/.agents/skills, so a later git pull keeps them current. In Codex, invoke a skill explicitly with $<skill-name> (for example $start).

After installing, start by setting up the product context and decision:

• "Set up product-eval for my product."
• "Help me decide what to build next."
• "Help me prepare a decision report."
• "Pressure-test this Q3 roadmap using the evidence in our current scope."

On first run, product-eval should ask for the product, target users, decision goal, scope/time horizon, available evidence sources, and the metric or business constraint that should shape the recommendation. It should not rank a vague area like "onboarding" or judge a bet like "SSO" until the product context and evidence plan are clear, unless you explicitly ask for an outside-in first pass with a lower confidence ceiling.

Each step should end with a Next move that points toward delivery: upload the evidence, rank the candidates, run the readiness call, pressure-test the plan, write the decision memo, render the scorecard, or hand a build brief to Codex/design.

60-second demo

Point it at your customer signal and ask for the outcome. For example, hand it a support-ticket export in Codex:

"These are our support tickets. Can you analyze them?"

product-eval runs the full chain (gather and weight evidence, synthesize themes, score Value and Confidence, run the readiness gate) and returns a ranked shortlist with a verdict on each.

Ask for a scorecard and it builds a self-contained, shareable HTML dashboard:

"Can you build a scorecard?"

No data of your own? It also works top-down from a domain ("research the observability tooling space and tell me what's worth building"), reasoning from public signal and stating the resulting confidence ceiling. You can also open a ready-made example with zero setup: examples/activation-q3/scorecard.html.

Outcome Modes

Rank Opportunities

Use this when you have a product area, evidence pile, or backlog and want a ranked shortlist. The plugin gathers or reads evidence, turns themes into buildable problem statements, scores each item by Value and Confidence, and returns a ranked table with the best next move.

Decide On One Bet