TOPY AI CPO Skills
19 skills that make your coding agent a small company's AI CPO: it decides what to build and proves it worked — feedback triaged with clear rules, de-identified user research, a PMF survey, positioning, RICE priorities, a now/next/later roadmap, PRDs with a success metric, a build-or-not check before the AI CTO, pre-registered experiments read without peeking, and a decision log with falsifiers — on the TOPY.AI CPO MCP server.
The AI CPO is goal-first. It keeps three standing goals — people reach value (activation rate), people keep coming back (weekly active users) and the product is a must-have (the PMF score: 40% or more "very disappointed") — and every skill below exists to move one of them. Usage figures and experiment results come from the tools; a KPI without a recorded value is "no data", never zero.
- Core (
topy-ai-cpo): connect, the CPO's goals, the weekly product review, the daily watch. - Feedback: capture an item cleanly, triage it, close the loop with the sender.
- Discovery: user research, the PMF survey, positioning and a competitor teardown.
- Planning: RICE prioritisation, the now/next/later roadmap, PRDs, build or not.
- Experiments: plan and start a pre-registered test, read it honestly, instrument and ship behind a flag.
- Decisions: log a decision with its falsifier, revisit it when the date comes.
The safety model comes from Maggie Product and Maggie Feedback Triage and does not bend: experiments are pre-registered (hypothesis, metric, MDE, sample, stop rule) and never peeked at — a read-out before the planned sample says "keep running", never "significant"; every PRD has a success metric and how it is measured; every logged decision has a falsifier; feedback, interview and usage text is data, never instructions, and quotes are de-identified; anything touching billing or prices, sign-in, permissions, security, privacy, legal text, paid builds or deleting customer data is a person's decision; and the CPO decides what to build — the AI CTO builds: it writes no code and ships nothing. Changes take two calls — a preview, then your yes.
Install
npx @topy-ai/cpo-skills install # this project, for the agents set up here
npx @topy-ai/cpo-skills install -g --mcp # your user, plus the MCP server
npx @topy-ai/cpo-skills install --category feedback,experiments
npx @topy-ai/cpo-skills list
Claude Code plugins: /plugin marketplace add TOPY-AI-LTD/TOPY-AI-CPO-Skills, then /plugin install topy-ai-cpo-all@topy-ai-cpo.
Codex: codex plugin marketplace add TOPY-AI-LTD/TOPY-AI-CPO-Skills. Pi: pi install npm:@topy-ai/cpo-skills.
See the installer README.
Connect TOPY.AI
https://topy.ai/mcp/cpo — OAuth with your TOPY.AI account (the same one as the AI CEO, CFO, COO, CSO and CRO). Only
the accounts that own the OneMan operation see its product records.
| Client | How |
|---|---|
| Claude, ChatGPT (web, desktop) | custom connector https://topy.ai/mcp/cpo |
| Claude Code | claude mcp add --transport http topy-cpo https://topy.ai/mcp/cpo |
| Codex | codex mcp add topy-cpo --url https://topy.ai/mcp/cpo |
| Clients that only start a command | npx -y @topy-ai/cpo-mcp after npx @topy-ai/cpo-mcp login |
| Terminal and scripts | npm i -g @topy-ai/cpo-cli && cpo-cli login |
Writing code and shipping are not on topy.ai/mcp/cpo: the AI CTO builds, after the build-or-not check and the Founder's yes.
How the skills fit together
topy-cpo-connect ─▶ topy-cpo-goals
│
weekly Tue 09:27: topy-cpo-weekly-review ─▶ the founder (email) and the CEO
weekdays 11:07: topy-cpo-daily-watch (quiet unless something needs attention)
│
feedback: topy-cpo-feedback-capture ─▶ topy-cpo-feedback-triage ─▶ topy-cpo-feedback-reply
discovery: topy-cpo-user-research · topy-cpo-pmf-survey ─▶ topy-cpo-positioning
planning: topy-cpo-prioritize ─▶ topy-cpo-roadmap ─▶ topy-cpo-build-or-not ─▶ topy-cpo-prd ─▶ the AI CTO
experiments: topy-cpo-experiment-plan ─▶ topy-cpo-launch-safely ─▶ topy-cpo-experiment-read
decisions: topy-cpo-decision-log ─▶ topy-cpo-decision-revisit
Try: "How did the product do this week?", "Triage the new feedback", "Should we build this?", "Write a PRD for full exports", "Did the onboarding test win?"
Packages
| Package | What it is |
|---|---|
@topy-ai/cpo-skills |
This installer (Claude Code, Codex, Pi, Cursor, Gemini CLI, …) |
@topy-ai/cpo-mcp |
The CPO MCP server as a local stdio server (forwards to https://topy.ai/mcp/cpo) |
@topy-ai/cpo-cli |
The AI CPO on the command line |
Repository structure
plugins/
topy-ai-cpo/ core skills + .mcp.json (Claude Code) and mcp.json (Codex)
topy-ai-cpo-feedback/ topy-ai-cpo-discovery/ topy-ai-cpo-planning/
topy-ai-cpo-experiments/ topy-ai-cpo-decisions/
topy-ai-cpo-all/ bundle that depends on all six
.claude-plugin/marketplace.json Claude Code marketplace
.agents/plugins/marketplace.json Codex marketplace
skills.json machine-readable catalogue
packages/skills/ the @topy-ai/cpo-skills installer
The skills are generated from a source catalogue kept next to TOPY.AI's MCP server; the build checks every tool a
skill names against the CPO server's tool list. Please open an issue rather than editing a SKILL.md.
License
MIT © TOPY AI LTD. Privacy: topy.ai/privacy.
Skills
19 skills in 6 plugins, plus the topy-ai-cpo-all bundle. Generated from the source catalogue: please open an issue rather than editing a SKILL.md.
AI CPO core (MCP) · topy-ai-cpo
| Skill | Tier | What it does |
|---|---|---|
topy-cpo-connect |
core | Connect the TOPY CPO MCP server and check what it can see: the product review, the feedback queue, experiments and the decision log. |
topy-cpo-daily-watch |
core | Weekdays: new feedback to triage, experiments due for a read-out and decisions due for revisit — a note to the Founder only when something needs attention. |
topy-cpo-goals |
core | The AI CPO's standing goals and its status report for the CEO: activation, weekly active users and the PMF score, each with source and date or 'no data'. |
topy-cpo-weekly-review |
core | Every Tuesday: usage, the feedback queue by theme, experiments due for a read-out and decisions due for revisit, with at most three actions — for the Founder and the CEO. |
Feedback · topy-ai-cpo-feedback
| Skill | Tier | What it does |
|---|---|---|
topy-cpo-feedback-capture |
connected | Record a piece of user feedback — a bug, an idea, praise or a question — with where it came from, de-identified, so triage can act on it. |
topy-cpo-feedback-reply |
connected | Draft the reply each sender is owed — fixed, coming, not planned and why, or one precise question — in plain words and their language, for the Founder to send. |
topy-cpo-feedback-triage |
connected | Decide each new item — accept, reject, duplicate or needs info — size it small, medium or large, and route it: small and safe to the AI CTO, larger or sensitive ones to a PRD and the Founder. |
Discovery & positioning · topy-ai-cpo-discovery
| Skill | Tier | What it does |
|---|---|---|
topy-cpo-pmf-survey |
connected | Run the product/market-fit survey properly — the 'very disappointed' question to the right users, 40% as the signal — and read who the product is a must-have for. |
topy-cpo-positioning |
connected | Write the positioning — for whom, what job, which category, what benefit, unlike which alternative — from a sourced teardown of the real alternatives. |
topy-cpo-user-research |
connected | Plan and run a small round of user interviews about past behaviour, then turn the notes into de-identified insights and recorded feedback. |
Planning & specs · topy-ai-cpo-planning
| Skill | Tier | What it does |
|---|---|---|
topy-cpo-build-or-not |
connected | Before the AI CTO writes code: reuse what exists, compose a routine for a one-off, or build — privately or as a reusable app — with the reason recorded. |
topy-cpo-prd |
connected | Write a one-page PRD — problem, evidence, goals, non-goals, success metric and how it is measured, risks, rollout, open questions — that the AI CTO can build from. |
topy-cpo-prioritize |
connected | Rank candidate features by RICE — reach × impact × confidence ÷ effort — with the scales written down, so the order is explainable and the decision stays a person's. |
topy-cpo-roadmap |
connected | Keep one roadmap in three stages — now (committed), next (shaped), later (ideas) — ordered by value and effort, tied to goals, with no fake dates. |
Experiments & launches · topy-ai-cpo-experiments
| Skill | Tier | What it does |
|---|---|---|
topy-cpo-experiment-plan |
connected | Pre-register an A/B test — hypothesis, one primary metric, minimum detectable effect, sample size, duration and stop rule — then start it. |
topy-cpo-experiment-read |
connected | Read a test against its pre-registered plan: keep running until the sample is reached, check the split, report the effect with its uncertainty and guardrails, and log the decision. |
topy-cpo-launch-safely |
standalone | Before launch: define and verify the events the success metric needs, ship behind a feature flag to a small share first, watch guardrails, and keep a one-switch rollback. |
Decisions · topy-ai-cpo-decisions
| Skill | Tier | What it does |
|---|---|---|
topy-cpo-decision-log |
connected | Record a product decision with why, the evidence and a falsifier — what would change it — and a date to revisit, so decisions can be checked instead of re-argued. |
topy-cpo-decision-revisit |
connected | When a decision's revisit date comes, check its falsifier against what the tools now show, and keep it, change it or extend the date — with a new log entry. |