pm-claude-skills 28.4.0 → 29.1.0

package/README.md

@@ -4,7 +4,7 @@
   </a>
 </p>
 
-# 🧠 PM Skills — 207 Professional Agent Skills for Claude, ChatGPT, Gemini, Cursor, Codex & Hermes
+# 🧠 PM Skills — 222 Professional Agent Skills for Claude, ChatGPT, Gemini, Cursor, Codex & Hermes
 
 [![Stars](https://img.shields.io/github/stars/mohitagw15856/pm-claude-skills?style=social)](https://github.com/mohitagw15856/pm-claude-skills/stargazers)
 [![npm](https://img.shields.io/npm/v/pm-claude-skills?logo=npm&color=cb3837)](https://www.npmjs.com/package/pm-claude-skills)
@@ -14,7 +14,7 @@
 [![MCP](https://img.shields.io/badge/MCP-registry-0098FF?logo=modelcontextprotocol&logoColor=white)](https://github.com/mohitagw15856/pm-claude-skills#-use-it-anywhere--the-ai-ecosystem)
 [![Remote MCP](https://img.shields.io/badge/ChatGPT%20%C2%B7%20Claude.ai-connector-10A37F)](mcp-remote/)
 [![Smithery](https://smithery.ai/badge/mohit15856/pm-skills)](https://smithery.ai/servers/mohit15856/pm-skills)
-[![Skills](https://img.shields.io/badge/skills-207-blue)](https://github.com/mohitagw15856/pm-claude-skills)
+[![Skills](https://img.shields.io/badge/skills-222-blue)](https://github.com/mohitagw15856/pm-claude-skills)
 [![eval](https://img.shields.io/badge/eval--scored-196%20%C2%B7%20avg%204.8%2F5-brightgreen)](https://mohitagw15856.github.io/pm-claude-skills/leaderboard.html)
 [![Subagents](https://img.shields.io/badge/subagents-4-blueviolet)](agents/)
 [![Commands](https://img.shields.io/badge/slash%20commands-6-blueviolet)](commands/)
@@ -30,22 +30,22 @@
 
 <p align="center">
   <a href="https://mohitagw15856.github.io/pm-claude-skills/">
-    <img src="https://readme-typing-svg.demolab.com?font=Inter&weight=600&size=24&pause=900&color=D97757&center=true&vCenter=true&width=720&lines=207+pro+AI+skills;PRDs%2C+roadmaps%2C+launches+%E2%80%94+done+right;Run+any+one+free+in+your+browser" alt="PM Skills" />
+    <img src="https://readme-typing-svg.demolab.com?font=Inter&weight=600&size=24&pause=900&color=D97757&center=true&vCenter=true&width=720&lines=222+pro+AI+skills;PRDs%2C+roadmaps%2C+launches+%E2%80%94+done+right;Run+any+one+free+in+your+browser" alt="PM Skills" />
   </a>
 </p>
 
-> **Generic AI gives you filler. These give you the structure a senior pro actually uses** — PRDs, exec updates, launch plans, postmortems — as open-source `SKILL.md` files. Across **21 professions**, not just product management. One source, every AI tool.
+> **Generic AI gives you filler. These give you the structure a senior pro actually uses** — PRDs, exec updates, launch plans, postmortems — as open-source `SKILL.md` files. Across **22 professions**, not just product management. One source, every AI tool.
 
 > **🆕 Now plugs into your stack** — automate skills in **[n8n](connectors/n8n.md)**, build apps on **[Lovable](connectors/lovable.md)**, and run them in your **[Obsidian](connectors/obsidian.md)** vault, all via a read-only **[REST API](mcp-remote/#rest-api-for-n8n-lovable-make-any-http-tool)** on the hosted Worker.
 
 ### ⭐ If this saves you time, [star the repo](https://github.com/mohitagw15856/pm-claude-skills) — it's the #1 way to help others find it.
 
 > **PM stands for Professional, not just Product Management.**
-> 207 professional skills + 4 agent templates across 28 bundles covering 21 professions. Built for Claude Code — and now portable to ChatGPT, Gemini, and Hermes Agent. Built by a PM, used by everyone.
+> 222 professional skills + 4 agent templates across 30 bundles covering 22 professions. Built for Claude Code — and now portable to ChatGPT, Gemini, and Hermes Agent. Built by a PM, used by everyone.
 
 A community-built library of professional skills for every field — product management, engineering, customer success, marketing, social media, writers, design, legal, finance, HR, sales, operations, research, and more. Each skill is a structured `SKILL.md` file that teaches an AI assistant how to produce professional-grade outputs for your workflows. Skills run natively in **Claude Code** and **Hermes Agent** (same open `SKILL.md` standard), and ship as ready-to-paste exports for **ChatGPT** and **Gemini** — see [Works With](#-works-with--cross-tool-compatibility).
 
-**🆕 Latest release (v28.4.0 — run it free, no credit card):** the playground no longer needs a paid key to start. Use a **free Google AI Studio key** (no card, ~30 seconds) or pick **In-browser** to run a model with **no key at all** (WebLLM on WebGPU — fully private, nothing leaves your device), and every skill now shows a **pre-generated sample output** so you can see the value with zero setup. Plus the **[Professional Brain is now in the browser](https://mohitagw15856.github.io/pm-claude-skills/brain.html)** — a 🧠 Brain page that feeds your durable memory into every run and saves outcomes back. *Earlier (v28.3 — the Professional Brain):* skills gained a durable, local markdown **memory** they read before answering, **write back** to (provenance-tagged, approval-gated), and even **act on** (open tickets/post updates via `action-runner`, dry-run + approval). Set it up in 5 minutes — **[Quickstart](BRAIN_QUICKSTART.md)** · [architecture](BRAIN.md). *Recent integrations:* **[n8n](connectors/n8n.md)**, **[Lovable](connectors/lovable.md)**, **[Obsidian](connectors/obsidian.md)** + a read-only **REST API** (`/v1`) on the hosted Worker. *Earlier (v28.0):* every channel is now shipped & public — **[`pip install pm-skills`](https://pypi.org/project/pm-skills/)** (LangChain/CrewAI), the **hosted MCP connector** `https://pm-skills-mcp.pm-claude-skills.workers.dev/` (add to **ChatGPT/Claude.ai/Cursor** by URL), and the **[official MCP registry](https://github.com/mohitagw15856/pm-claude-skills/blob/main/server.json)** listing. See the new **[🌍 Use It Anywhere](#-use-it-anywhere--the-ai-ecosystem)** section. Also: **multi-model playground** (run any skill with your own **Claude / OpenAI / Gemini / Ollama** key — local models included), a **[remote MCP connector](mcp-remote/)** (add PM Skills to **ChatGPT, Claude.ai & Cursor** with one URL — no install), the MCP server now exposes skills as **prompts + resources**, **4 new IDE targets** (Cline, Continue, Zed, Roo — 9 platforms total), a **[Python package](python/)** (`pip install pm-skills` → LangChain/CrewAI tools), an **[`llms.txt`](https://mohitagw15856.github.io/pm-claude-skills/llms.txt)** for agent discovery, a reusable **[score-skills GitHub Action](.github/actions/score-skills/)**, and **MCP-registry manifests**. Plus the full library is now **eval-scored: 196 skills, avg 4.8/5**. **207 skills**. See the [changelog](#-changelog).
+**🆕 Latest release (v29.1.0 — 15 new skills + composable recipes):** two new bundles ship the library's biggest gaps — **AI/ML Engineering** (`model-card`, `ai-eval-plan`, `rag-design-doc`, `agent-spec`, `dataset-datasheet`, `ai-feature-prd`, `llm-cost-latency-budget`) and **Growth/Marketing** (`marketing-funnel-plan`, `lifecycle-crm-plan`, `paid-acquisition-plan`, `growth-experiment-backlog`, `retention-loop-design`, `referral-program-design`, `messaging-framework`, `conversion-rate-optimization`). Plus **two new [Workflow Recipes](WORKFLOWS.md)** that chain them end-to-end ("Launch an AI Feature", "Grow a Product") — now **loadable right in the [Workflow Canvas](https://mohitagw15856.github.io/pm-claude-skills/canvas.html)** and **Brain-aware**. The memory+action moat also deepened: **18 skills are now brain-aware** and `action-runner` gained explicit **GitHub / Linear / Slack / Notion / Email / Calendar** targets with per-action risk gating. *Earlier (v29.0 — Multilingual + a polished experience):* run any skill in **10 languages** (Spanish, Mandarin, Hindi, Arabic — with RTL — Portuguese, French, German, Japanese, Russian, Indonesian); the frameworks are universal, so the model localizes the whole output. Plus a **[15-page Practical Guide](web/guide.html)** and a **[one-page cheatsheet](CHEATSHEET.md)** (PDF/PNG), a **navigation + playground UX revamp** (a repo link from the playground, sticky nav, a redesigned home), and quality hardening — **[case studies](CASE_STUDIES.md)**, opt-in 👍/👎 feedback, and a CI **regression gate** that blocks score drops. *Earlier (v28.4 — run it free, no credit card):* the playground no longer needs a paid key to start. Use a **free Google AI Studio key** (no card, ~30 seconds) or pick **In-browser** to run a model with **no key at all** (WebLLM on WebGPU — fully private, nothing leaves your device), and every skill now shows a **pre-generated sample output** so you can see the value with zero setup. Plus the **[Professional Brain is now in the browser](https://mohitagw15856.github.io/pm-claude-skills/brain.html)** — a 🧠 Brain page that feeds your durable memory into every run and saves outcomes back. *Earlier (v28.3 — the Professional Brain):* skills gained a durable, local markdown **memory** they read before answering, **write back** to (provenance-tagged, approval-gated), and even **act on** (open tickets/post updates via `action-runner`, dry-run + approval). Set it up in 5 minutes — **[Quickstart](BRAIN_QUICKSTART.md)** · [architecture](BRAIN.md). *Recent integrations:* **[n8n](connectors/n8n.md)**, **[Lovable](connectors/lovable.md)**, **[Obsidian](connectors/obsidian.md)** + a read-only **REST API** (`/v1`) on the hosted Worker. *Earlier (v28.0):* every channel is now shipped & public — **[`pip install pm-skills`](https://pypi.org/project/pm-skills/)** (LangChain/CrewAI), the **hosted MCP connector** `https://pm-skills-mcp.pm-claude-skills.workers.dev/` (add to **ChatGPT/Claude.ai/Cursor** by URL), and the **[official MCP registry](https://github.com/mohitagw15856/pm-claude-skills/blob/main/server.json)** listing. See the new **[🌍 Use It Anywhere](#-use-it-anywhere--the-ai-ecosystem)** section. Also: **multi-model playground** (run any skill with your own **Claude / OpenAI / Gemini / Ollama** key — local models included), a **[remote MCP connector](mcp-remote/)** (add PM Skills to **ChatGPT, Claude.ai & Cursor** with one URL — no install), the MCP server now exposes skills as **prompts + resources**, **4 new IDE targets** (Cline, Continue, Zed, Roo — 9 platforms total), a **[Python package](python/)** (`pip install pm-skills` → LangChain/CrewAI tools), an **[`llms.txt`](https://mohitagw15856.github.io/pm-claude-skills/llms.txt)** for agent discovery, a reusable **[score-skills GitHub Action](.github/actions/score-skills/)**, and **MCP-registry manifests**. Plus the full library is now **eval-scored: 196 skills, avg 4.8/5**. **222 skills**. See the [changelog](#-changelog).
 
 ### ▶ See it in action — [try the live Skill Playground](https://mohitagw15856.github.io/pm-claude-skills/)
 
@@ -74,13 +74,43 @@ A community-built library of professional skills for every field — product man
 | 🔭 [Competitor Teardown](skills/competitor-teardown) | "what are rivals up to?" | a positioning map, feature gaps & strategy |
 | 📝 [Meeting Notes](skills/meeting-notes) | a raw transcript | decisions, owners & next steps |
 
-→ Want proof first? See [**real sample outputs**](https://mohitagw15856.github.io/pm-claude-skills/examples.html) from each skill. Like what you see? [**Install in 2 minutes**](#-quick-install-2-minutes) · [browse all 207 skills](#️-all-207-skills) · [**⭐ star the repo**](https://github.com/mohitagw15856/pm-claude-skills/stargazers) so others find it.
+→ Want proof first? See [**real sample outputs**](https://mohitagw15856.github.io/pm-claude-skills/examples.html) from each skill. Like what you see? [**Install in 2 minutes**](#-quick-install-2-minutes) · [browse all 222 skills](#️-all-222-skills) · [**⭐ star the repo**](https://github.com/mohitagw15856/pm-claude-skills/stargazers) so others find it.
+
+---
+
+## 📄 One-page cheatsheet
+
+The whole library on one page — install, all 222 skills by profession, the Professional Brain, the REST API, integrations, and quality tooling. Great to **print, share, or drop into a slide**.
+
+<p align="center">
+  <a href="CHEATSHEET.md">
+    <img src="web/docs-assets/cheatsheet.png" width="80%" alt="PM Skills one-page cheatsheet — 222 skills, the Professional Brain, REST API, integrations, and tooling at a glance." />
+  </a>
+</p>
+
+**[📥 Markdown](CHEATSHEET.md)** · **[🖼️ PNG](web/docs-assets/cheatsheet.png)** · **[📄 PDF](web/docs-assets/cheatsheet.pdf)** — regenerate the image/PDF with `node web/docs-assets/shoot.mjs` (`PAGES=cheatsheet`).
+
+> 📖 **Want the full walkthrough?** The **[Practical Guide](web/guide.html)** is a 15-page,
+> example-driven manual — install → your first skill → worked examples (PRD, exec update, RICE,
+> postmortem) → chaining recipes → the Brain → running anywhere → tips. **[Download the PDF →](web/docs-assets/PM-Skills-Guide.pdf)**
+
+---
+
+## 🧠 The Professional Brain — local-first memory for any AI agent
+
+Generic AI forgets everything between sessions, so you re-paste context forever and last quarter's *why* evaporates. The **Professional Brain** is the fix — and it's the most novel piece here, worth a look on its own:
+
+- **Plain markdown, no vector DB** — a `brain/` folder (knowledge · decisions · hypotheses · stakeholders) your AI **reads before answering and writes to after**. Grep-able, auditable, Obsidian-compatible — memory you can read and correct by hand, not a black box.
+- **Provenance over confidence theatre** — every fact is tagged `[data] [interview] [external] [verbal] [hunch]`, so a hunch never poses as a measured result.
+- **It acts, safely** — the [`action-runner`](skills/action-runner/) turns a skill's recommendations into real tickets/messages: dry-run, risk-rated, approval-gated, then recorded back. *Nothing acts silently.*
+
+**→ [Read the architecture](BRAIN.md)** · **[5-min Quickstart](BRAIN_QUICKSTART.md)** · **[try it in the browser](https://mohitagw15856.github.io/pm-claude-skills/brain.html)**
 
 ---
 
 ## 🔄 One library, the whole professional workflow
 
-These 207 skills aren't a random catalog — they cover the **full arc of professional work**, end to end. Wherever you are in the loop, there's a skill for it:
+These 222 skills aren't a random catalog — they cover the **full arc of professional work**, end to end. Wherever you are in the loop, there's a skill for it:
 
 ```mermaid
 flowchart LR
@@ -99,7 +129,7 @@ flowchart LR
 | **📊 Measure** | Track outcomes & analyse | `metrics-framework` · `cohort-analysis` · `ab-test-planner` · `churn-analysis` |
 | **📣 Communicate** | Report up and out | `executive-update` · `board-deck-narrative` · `stakeholder-update` · `qbr-deck` |
 
-> New here? Start with the [**top-tier skills**](#️-skill-tiers--start-with-the-strongest), or jump straight to [**all 207 skills**](#️-all-207-skills) grouped by profession.
+> New here? Start with the [**top-tier skills**](#️-skill-tiers--start-with-the-strongest), or jump straight to [**all 222 skills**](#️-all-222-skills) grouped by profession.
 
 ---
 
@@ -167,7 +197,9 @@ The bot runs the skill and posts the result as a reply. `/skill list` shows usag
 
 **Grounded in canonical frameworks.** These aren't invented prompts — each skill encodes a proven method and cites it: RICE (Intercom), Jobs-to-be-Done (Christensen), Continuous Discovery (Teresa Torres), Porter's Five Forces, the Pyramid Principle (Minto), Google SRE, WCAG, *Obviously Awesome* (April Dunford), and more. The source shows as a **"📚 Based on"** line on every [skill page](https://mohitagw15856.github.io/pm-claude-skills/skill/rice-prioritisation.html) and in the Playground.
 
-**And measured, not just claimed.** An [eval harness](evals/) runs each skill against a held-out test case, then an LLM judge (Opus 4.8) rates the output on four dimensions — **structure, completeness, usefulness, grounding** — averaged across two models. **15 skills are eval-scored** today (and climbing); the rest are reviewed against the [authoring standard](SKILL-AUTHORING-STANDARD.md).
+**And measured, not just claimed.** An [eval harness](evals/) runs each skill against a held-out test case, then an LLM judge (Opus 4.8) rates the output on four dimensions — **structure, completeness, usefulness, grounding** — averaged across two models. **15 skills are eval-scored** today (and climbing); the rest are reviewed against the [authoring standard](SKILL-AUTHORING-STANDARD.md). A **[regression gate](.github/workflows/skill-pr-check.yml)** then blocks any PR that drops a skill's score, so quality can't quietly rot as models change.
+
+**The qualitative half — real time saved.** Scores measure quality; **[case studies](CASE_STUDIES.md)** measure outcomes. See the kind of before → after a skill delivers (PRD: ~2 days → ~25 min), and **[add your own 2-minute ROI story](../../issues/new?template=roi-story.yml)** — we'll feature it.
 
 **The loop actually catches bad skills.** A recent run flagged three skills scoring ~2/5 because they asked for missing inputs instead of delivering. We added a "work from a brief" rule, re-ran, and they jumped to **4.75/5**:
 
@@ -195,7 +227,7 @@ Most skill repos are a folder of prompts. This one is a **system** — measured,
 
 | | **PM Skills** | Typical skill repo |
 |---|:---:|:---:|
-| Skills | **207**, across 21 professions | a handful → dozens, usually one domain |
+| Skills | **222**, across 22 professions | a handful → dozens, usually one domain |
 | Quality | **eval-scored** on a rubric + a public [benchmark](https://mohitagw15856.github.io/pm-claude-skills/benchmark.html) | trust the README |
 | Improves itself | ✅ eval → critique → rewrite (kept only if it scores higher) | ✗ |
 | Grounded in frameworks | ✅ each cites its source (RICE, JTBD, Pyramid Principle…) | rarely |
@@ -220,7 +252,7 @@ Most skill repos are a folder of prompts. This one is a **system** — measured,
 - [📦 Plugin Directory](#-plugin-directory)
 - [🤖 Building Blocks for Agent Templates](#-building-blocks-for-agent-templates)
 - [🏷️ Skill Tiers — start with the strongest](#️-skill-tiers--start-with-the-strongest)
-- [🗂️ All 207 Skills](#️-all-207-skills)
+- [🗂️ All 222 Skills](#️-all-222-skills)
 - [📋 Changelog](#-changelog)
 - [🤝 Contributing](#-contributing--add-your-skill)
 - [🔗 Related Projects](#-related-projects)
@@ -229,13 +261,13 @@ Most skill repos are a folder of prompts. This one is a **system** — measured,
 
 ## 🚀 Quick Install (2 minutes)
 
-**With the [`skills`](https://github.com/vercel-labs/skills) CLI** (the open agent-skills installer that works across Claude Code, Cursor, Codex, OpenCode & 60+ agents) — pick from all 207 interactively:
+**With the [`skills`](https://github.com/vercel-labs/skills) CLI** (the open agent-skills installer that works across Claude Code, Cursor, Codex, OpenCode & 60+ agents) — pick from all 222 interactively:
 
 ```bash
 npx skills add mohitagw15856/pm-claude-skills            # browse & pick (auto-detects your agent)
 npx skills add mohitagw15856/pm-claude-skills --list     # just preview the catalog
 npx skills add mohitagw15856/pm-claude-skills --skill prd-template   # grab one
-npx skills add mohitagw15856/pm-claude-skills --skill '*'            # install all 207
+npx skills add mohitagw15856/pm-claude-skills --skill '*'            # install all 222
 ```
 
 **Or our own installer** — via the [`pm-claude-skills`](https://www.npmjs.com/package/pm-claude-skills) npm package (Windows/macOS/Linux, needs Node), which also installs subagents, slash commands & cross-tool exports:
@@ -244,7 +276,7 @@ npx skills add mohitagw15856/pm-claude-skills --skill '*'            # install a
 npx pm-claude-skills add --agent claude     # or: codex · cursor · hermes · openclaw
 ```
 
-**Or one-line MCP** — make all 207 skills + 5 workflow recipes available in *every* session of any MCP client (Claude Code, Claude Desktop, Cursor, Windsurf), no per-file install:
+**Or one-line MCP** — make all 222 skills + 5 workflow recipes available in *every* session of any MCP client (Claude Code, Claude Desktop, Cursor, Windsurf), no per-file install:
 
 ```bash
 claude mcp add pm-skills -- npx -y pm-claude-skills-mcp
@@ -301,7 +333,7 @@ ln -s ~/pm-claude-skills/skills/* ~/.claude/skills/
 
 ## 🌍 Use It Anywhere — the AI Ecosystem
 
-The same 207 skills reach you through every channel — pick whatever fits your stack:
+The same 222 skills reach you through every channel — pick whatever fits your stack:
 
 | Channel | Get it |
 |---|---|
@@ -315,7 +347,7 @@ The same 207 skills reach you through every channel — pick whatever fits your
 | 🖥️ **IDE rules** | Generated exports for **Cursor, Windsurf, Aider, Cline, Continue, Zed, Roo** — [`exports/`](exports/) |
 | 🤖 **Agents & answer engines** | [`llms.txt`](https://mohitagw15856.github.io/pm-claude-skills/llms.txt) makes the whole library discoverable & citable |
 
-> **196 of 207 skills are eval-scored, averaging 4.8/5** — see the [leaderboard](https://mohitagw15856.github.io/pm-claude-skills/leaderboard.html).
+> **196 of 222 skills are eval-scored, averaging 4.8/5** — see the [leaderboard](https://mohitagw15856.github.io/pm-claude-skills/leaderboard.html).
 
 > 🧠 **Give your skills memory:** the [Professional Brain](BRAIN.md) is a local markdown folder skills read, write back to, and act on. It's a folder + one file to start — **[5-minute Quickstart](BRAIN_QUICKSTART.md)**.
 
@@ -332,7 +364,7 @@ There are two kinds of support. **Native `SKILL.md` agents** read the file as-is
 auto-discover skills from the `description` frontmatter. **Other tools** take the markdown
 body as a system prompt — for those we ship ready-made [exports](#ready-to-use-exports).
 
-**In your editor (VS Code / Cursor):** the [**`vscode-extension/`**](vscode-extension/) brings all 207 skills into the Command Palette — search and *insert a skill as context* for Copilot/Cursor chat, copy it, or open it in the Playground.
+**In your editor (VS Code / Cursor):** the [**`vscode-extension/`**](vscode-extension/) brings all 222 skills into the Command Palette — search and *insert a skill as context* for Copilot/Cursor chat, copy it, or open it in the Playground.
 
 | Platform | How it works | Auto-trigger? |
 |---|---|---|
@@ -477,7 +509,7 @@ ANTHROPIC_API_KEY=sk-ant-… npx pm-claude-skills generate --from ./team-process
 
 **🏆 Skill Leaderboard — [evals](evals/).** An LLM-as-judge harness scores each skill across Claude models on structure, completeness, usefulness, and grounding. **[View the leaderboard →](https://mohitagw15856.github.io/pm-claude-skills/leaderboard.html)**
 
-> **Score the *whole* library, cheaply.** `npm run eval:gen-cases` writes a representative input for **all 207 skills** (curated cases kept verbatim; the rest auto-generated), then `npm run eval:all` runs the full pass on Haiku for **~$2** total. The weakest scorers can then be auto-rewritten with **`npm run improve -- <skill>`** (`/improve`), which keeps a change only if the score goes up.
+> **Score the *whole* library, cheaply.** `npm run eval:gen-cases` writes a representative input for **all 222 skills** (curated cases kept verbatim; the rest auto-generated), then `npm run eval:all` runs the full pass on Haiku for **~$2** total. The weakest scorers can then be auto-rewritten with **`npm run improve -- <skill>`** (`/improve`), which keeps a change only if the score goes up.
 
 ---
 
@@ -553,7 +585,7 @@ Not sure which plugin to install? Here's what each one covers:
 
 On May 5, 2026, Anthropic [released their first agent templates](https://www.anthropic.com/news/finance-agents) — pre-packaged Claude agents that combine **skills, connectors, and subagents** into ready-to-run workflows for financial services.
 
-This library is the largest open-source collection of professional skills available — covering 20 professions beyond financial services. **The 207 skills here are the building blocks for agent templates outside of finance.**
+This library is the largest open-source collection of professional skills available — covering 20 professions beyond financial services. **The 222 skills here are the building blocks for agent templates outside of finance.**
 
 ### What is an agent template?
 
@@ -632,11 +664,11 @@ More templates will follow. If you want to contribute one, see the [template con
 
 ## 📋 Changelog
 
-**Latest: v28.1.0 — Integrations (n8n, Lovable, Obsidian) + a `/v1` REST API.** PM Skills now runs inside [n8n](connectors/n8n.md), powers [Lovable](connectors/lovable.md) apps, and lives in your [Obsidian](connectors/obsidian.md) vault — all backed by a read-only REST API on the hosted Worker (auto-deployed via a new GitHub Action). Builds on **v28.0.0 — live across the whole AI ecosystem.** Now public on every channel: **PyPI** (`pip install pm-skills`), a **hosted MCP connector** URL for ChatGPT/Claude.ai/Cursor, and the **official MCP registry** — plus a new **🌍 Use It Anywhere** section + PyPI/MCP header badges. Builds on the v27 ecosystem work: multi-model playground (Claude / OpenAI / Gemini / **Ollama**); a **remote MCP connector** ([`mcp-remote/`](mcp-remote/)) to add PM Skills to **ChatGPT, Claude.ai & Cursor** by URL; MCP server now serves skills as **prompts + resources**; **4 new IDE export targets** (Cline, Continue, Zed, Roo → 9 platforms); a **Python package** ([`python/`](python/), `pip install pm-skills` with LangChain/CrewAI adapters); an **`llms.txt`** for agent discovery; a reusable **[score-skills Action](.github/actions/score-skills/)**; and **MCP-registry manifests** (`server.json`, `smithery.yaml`). The full library is now eval-scored — **196 skills, avg 4.8/5**. Builds on **v26.0.0** (the Creator stack). **207 skills** across **28 bundles**, **21 professions**.
+**Latest: v28.1.0 — Integrations (n8n, Lovable, Obsidian) + a `/v1` REST API.** PM Skills now runs inside [n8n](connectors/n8n.md), powers [Lovable](connectors/lovable.md) apps, and lives in your [Obsidian](connectors/obsidian.md) vault — all backed by a read-only REST API on the hosted Worker (auto-deployed via a new GitHub Action). Builds on **v28.0.0 — live across the whole AI ecosystem.** Now public on every channel: **PyPI** (`pip install pm-skills`), a **hosted MCP connector** URL for ChatGPT/Claude.ai/Cursor, and the **official MCP registry** — plus a new **🌍 Use It Anywhere** section + PyPI/MCP header badges. Builds on the v27 ecosystem work: multi-model playground (Claude / OpenAI / Gemini / **Ollama**); a **remote MCP connector** ([`mcp-remote/`](mcp-remote/)) to add PM Skills to **ChatGPT, Claude.ai & Cursor** by URL; MCP server now serves skills as **prompts + resources**; **4 new IDE export targets** (Cline, Continue, Zed, Roo → 9 platforms); a **Python package** ([`python/`](python/), `pip install pm-skills` with LangChain/CrewAI adapters); an **`llms.txt`** for agent discovery; a reusable **[score-skills Action](.github/actions/score-skills/)**; and **MCP-registry manifests** (`server.json`, `smithery.yaml`). The full library is now eval-scored — **196 skills, avg 4.8/5**. Builds on **v26.0.0** (the Creator stack). **222 skills** across **30 bundles**, **22 professions**.
 
 Full [Keep a Changelog](https://keepachangelog.com/)-format history — every release back to the start — is in **[CHANGELOG.md](CHANGELOG.md)**.
 
-→ Earlier releases (v20 and before — the road from 6 to 207 skills) are in **[CHANGELOG.md](CHANGELOG.md)**.
+→ Earlier releases (v20 and before — the road from 6 to 222 skills) are in **[CHANGELOG.md](CHANGELOG.md)**.
 
 ---
 
@@ -672,7 +704,7 @@ This repo was built alongside a published 16-part article series on Medium.
 
 ## 🏷️ Skill Tiers — Start With the Strongest
 
-A 207-skill library doesn't have 207 equally-mature skills, and pretending otherwise
+A 222-skill library doesn't have 222 equally-mature skills, and pretending otherwise
 wastes your time. Skills are tiered honestly so you can start with the best work:
 
 - 🟢 **Production-Ready (50)** — battle-tested, stable output, used in real work. Includes the three skills with computed Python helpers (sprint planning, RICE, customer health). **Start here.**
@@ -685,7 +717,7 @@ If you're new, install `pm-essentials` and try a couple of Production-Ready skil
 
 ---
 
-## 🗂️ All 207 Skills
+## 🗂️ All 222 Skills
 
 Every skill, grouped by profession. **[Browse the full per-skill catalog → SKILLS.md](SKILLS.md)** · **[searchable live catalog](https://mohitagw15856.github.io/pm-claude-skills/catalog.html)** · **[run any skill in the browser](https://mohitagw15856.github.io/pm-claude-skills/)**
 
@@ -719,7 +751,7 @@ Every skill, grouped by profession. **[Browse the full per-skill catalog → SKI
 
 ## ❤️ Sponsor This Work
 
-Building and maintaining 207 skills across 28 bundles takes real time — testing skills against new model releases, building new ones from community requests, writing the article series, and keeping documentation current.
+Building and maintaining 222 skills across 30 bundles takes real time — testing skills against new model releases, building new ones from community requests, writing the article series, and keeping documentation current.
 
 If these skills save you time at work — or you're a company that wants your logo in front of the PMs, engineers, and operators who use them daily — **[become a sponsor →](https://github.com/sponsors/mohitagw15856)** (or [☕ buy me a coffee](https://www.buymeacoffee.com/mohit15856)).
 

package/commands/grow-a-product.md

@@ -0,0 +1,17 @@
+---
+description: Workflow recipe — turn a growth goal into a funnel diagnosis, experiment backlog, retention loop, and lifecycle journeys by chaining 4 skills.
+argument-hint: [the growth goal — e.g. "lift D30 retention for B2B teams"]
+---
+
+Run the **Grow a Product** workflow recipe for: $ARGUMENTS
+
+This is a *chain* of skills. Run each stage in order and **carry every stage's output forward as context** for the next — that shared context is the whole point. Open with a one-line plan of the 4 stages, then ask once for any essential missing inputs (the product & motion, current funnel numbers if any, the goal). Don't re-ask between stages.
+
+Run each stage under a clear `## Stage N — <name>` heading:
+
+1. **Diagnose the funnel** — apply the `marketing-funnel-plan` skill to map the full funnel, identify the single biggest leak, and set a 90-day focus stage and metric.
+2. **Build the experiment backlog** — apply the `growth-experiment-backlog` skill to turn that focus into prioritised, properly-powered hypotheses (ICE) with test designs.
+3. **Design the retention loop** — apply the `retention-loop-design` skill to design the engagement loop (trigger → action → reward → investment) and the activation→habit path that keeps the gains.
+4. **Nurture with lifecycle journeys** — apply the `lifecycle-crm-plan` skill to design behaviour-triggered journeys that drive the loop, with segmentation, holdouts, and suppression.
+
+Do not invent metrics or numbers — note assumptions instead. After the last stage, end with a 4-bullet **"What you now have"** recap linking each artifact to the stage that produced it.

package/commands/launch-an-ai-feature.md

@@ -0,0 +1,18 @@
+---
+description: Workflow recipe — take an AI/LLM feature from a probabilistic-aware PRD to a launch-ready model card by chaining 5 skills.
+argument-hint: [the AI feature idea]
+---
+
+Run the **Launch an AI Feature** workflow recipe for: $ARGUMENTS
+
+This is a *chain* of skills. Run each stage in order and **carry every stage's output forward as context** for the next — that shared context is the whole point. Open with a one-line plan of the 5 stages, then ask once for any essential missing inputs (the user problem, the stakes / cost of a wrong answer, the data available). Don't re-ask between stages.
+
+Run each stage under a clear `## Stage N — <name>` heading:
+
+1. **Spec it** — apply the `ai-feature-prd` skill to turn the idea into a PRD built for a probabilistic system: the UX of uncertainty, guardrails, fallback behaviour, and an explicit quality bar tied to the stakes.
+2. **Design the system** — apply the `rag-design-doc` skill (or note if an agent is the better fit — see `agent-spec`) to design retrieval/generation: chunking, retrieval, reranking, grounded answers, and the failure-mode table.
+3. **Plan evaluation** — apply the `ai-eval-plan` skill to define datasets, rubrics, baselines, the explicit ship threshold, and the regression gate.
+4. **Budget cost & latency** — apply the `llm-cost-latency-budget` skill for per-request token math, model tiering, caching, p95 targets, and spend guardrails.
+5. **Document it** — apply the `model-card` skill to produce a launch-ready card: intended use, sliced evaluation, limitations, and a rollback trigger.
+
+Do not invent metrics, costs, or eval results — note assumptions instead. After the last stage, end with a 5-bullet **"What you now have"** recap linking each artifact to the stage that produced it.

package/exports/README.md

@@ -8,7 +8,7 @@ by hand; edit the source skill and run:
 node scripts/build-exports.mjs
 ```
 
-Currently exporting **207 skills** to:
+Currently exporting **222 skills** to:
 
 - **ChatGPT — Custom GPT instructions** → `exports/chatgpt/`
 - **Google Gemini — Gem instructions** → `exports/gemini/`

package/exports/aider/README.md

@@ -3,7 +3,7 @@
 > Auto-generated from `skills/*/SKILL.md` by `scripts/build-exports.mjs`.
 > **Do not edit these files by hand** — edit the source skill and regenerate.
 
-207 skills exported. Copy a `.mdc rule` into the tool to use it.
+222 skills exported. Copy a `.mdc rule` into the tool to use it.
 
 | Skill | Bundle | Path |
 |---|---|---|
@@ -14,7 +14,10 @@
 | Account Plan | `pm-sales` | `pm-sales/account-plan/account-plan.md` |
 | Action Runner | `pm-cross` | `pm-cross/action-runner/action-runner.md` |
 | AEO Optimizer | `pm-writers` | `pm-writers/aeo-optimizer/aeo-optimizer.md` |
+| Agent Spec | `pm-ai` | `pm-ai/agent-spec/agent-spec.md` |
 | AI Ethics Review | `pm-advanced` | `pm-advanced/ai-ethics-review/ai-ethics-review.md` |
+| AI Eval Plan | `pm-ai` | `pm-ai/ai-eval-plan/ai-eval-plan.md` |
+| AI Feature PRD | `pm-ai` | `pm-ai/ai-feature-prd/ai-feature-prd.md` |
 | AI Product Canvas | `pm-advanced` | `pm-advanced/ai-product-canvas/ai-product-canvas.md` |
 | Ambiguity Resolver | `pm-strategy` | `pm-strategy/ambiguity-resolver/ambiguity-resolver.md` |
 | API Docs Writer | `pm-engineering` | `pm-engineering/api-docs-writer/api-docs-writer.md` |
@@ -46,6 +49,7 @@
 | Content Repurposer | `pm-creator` | `pm-creator/content-repurposer/content-repurposer.md` |
 | Context Mode | `pm-engineering` | `pm-engineering/context-mode/context-mode.md` |
 | Contract Review | `pm-legal` | `pm-legal/contract-review/contract-review.md` |
+| Conversion Rate Optimization | `pm-growth` | `pm-growth/conversion-rate-optimization/conversion-rate-optimization.md` |
 | Creator Brand Kit | `pm-creator` | `pm-creator/creator-brand-kit/creator-brand-kit.md` |
 | Creator Media Kit | `pm-creator` | `pm-creator/creator-media-kit/creator-media-kit.md` |
 | Customer Escalation Brief | `pm-cs` | `pm-cs/cs-escalation-brief/cs-escalation-brief.md` |
@@ -58,6 +62,7 @@
 | Data Quality Audit | `pm-data` | `pm-data/data-quality-audit/data-quality-audit.md` |
 | Database Migration Plan | `pm-engineering` | `pm-engineering/database-migration-plan/database-migration-plan.md` |
 | Database Schema Design | `pm-engineering` | `pm-engineering/database-schema-design/database-schema-design.md` |
+| Dataset Datasheet | `pm-ai` | `pm-ai/dataset-datasheet/dataset-datasheet.md` |
 | Debugging Log Analyser | `pm-engineering` | `pm-engineering/debugging-log-analyser/debugging-log-analyser.md` |
 | Demand Letter | `pm-legal` | `pm-legal/demand-letter/demand-letter.md` |
 | Dependency Audit | `pm-engineering` | `pm-engineering/dependency-audit/dependency-audit.md` |
@@ -99,6 +104,7 @@
 | Go-To-Market | `pm-gtm` | `pm-gtm/go-to-market/go-to-market.md` |
 | Go-to-Market Planner | `pm-delivery` | `pm-delivery/go-to-market-planner/go-to-market-planner.md` |
 | Grant Proposal | `pm-cross` | `pm-cross/grant-proposal/grant-proposal.md` |
+| Growth Experiment Backlog | `pm-growth` | `pm-growth/growth-experiment-backlog/growth-experiment-backlog.md` |
 | Hiring Rubric | `pm-people` | `pm-people/hiring-rubric/hiring-rubric.md` |
 | Hook Writer | `pm-creator` | `pm-creator/hook-writer/hook-writer.md` |
 | IEP Goal Support | `pm-education` | `pm-education/iep-goal-support/iep-goal-support.md` |
@@ -116,14 +122,19 @@
 | Launch Readiness | `pm-delivery` | `pm-delivery/launch-readiness/launch-readiness.md` |
 | Legal Brief | `pm-legal` | `pm-legal/legal-brief/legal-brief.md` |
 | Lesson Plan | `pm-education` | `pm-education/lesson-plan/lesson-plan.md` |
+| Lifecycle / CRM Plan | `pm-growth` | `pm-growth/lifecycle-crm-plan/lifecycle-crm-plan.md` |
 | Literature Review | `pm-research` | `pm-research/literature-review/literature-review.md` |
+| LLM Cost & Latency Budget | `pm-ai` | `pm-ai/llm-cost-latency-budget/llm-cost-latency-budget.md` |
 | Load Testing Plan | `pm-engineering` | `pm-engineering/load-testing-plan/load-testing-plan.md` |
 | Local Dev Setup | `pm-engineering` | `pm-engineering/local-dev-setup/local-dev-setup.md` |
+| Marketing Funnel Plan | `pm-growth` | `pm-growth/marketing-funnel-plan/marketing-funnel-plan.md` |
 | Media Pitch | `pm-gtm` | `pm-gtm/media-pitch/media-pitch.md` |
 | Meeting Notes | `pm-essentials` | `pm-essentials/meeting-notes/meeting-notes.md` |
+| Messaging Framework | `pm-growth` | `pm-growth/messaging-framework/messaging-framework.md` |
 | Metric Tree Builder | `pm-data` | `pm-data/metric-tree-builder/metric-tree-builder.md` |
 | Metrics Framework | `pm-data` | `pm-data/metrics-framework/metrics-framework.md` |
 | Microservices Decomposition | `pm-engineering` | `pm-engineering/microservices-decomposition/microservices-decomposition.md` |
+| Model Card | `pm-ai` | `pm-ai/model-card/model-card.md` |
 | Monitoring Setup Guide | `pm-engineering` | `pm-engineering/monitoring-setup-guide/monitoring-setup-guide.md` |
 | Morning Intelligence | `pm-operations` | `pm-operations/morning-intelligence/morning-intelligence.md` |
 | Multi-Source Signal Synthesiser | `pm-advanced` | `pm-advanced/multi-source-signal-synthesiser/multi-source-signal-synthesiser.md` |
@@ -134,6 +145,7 @@
 | OKR Builder | `pm-planning` | `pm-planning/okr-builder/okr-builder.md` |
 | Onboarding Plan | `pm-hr` | `pm-hr/onboarding-plan/onboarding-plan.md` |
 | On-Call Runbook | `pm-engineering` | `pm-engineering/oncall-runbook/oncall-runbook.md` |
+| Paid Acquisition Plan | `pm-growth` | `pm-growth/paid-acquisition-plan/paid-acquisition-plan.md` |
 | Parent Communication | `pm-education` | `pm-education/parent-communication/parent-communication.md` |
 | Partnership Proposal | `pm-sales` | `pm-sales/partnership-proposal/partnership-proposal.md` |
 | Patient Communication | `pm-research` | `pm-research/patient-communication/patient-communication.md` |
@@ -156,12 +168,15 @@
 | QBR Deck | `pm-cs` | `pm-cs/qbr-deck/qbr-deck.md` |
 | Quiz Generator | `pm-education` | `pm-education/quiz-generator/quiz-generator.md` |
 | RACI Matrix | `pm-operations` | `pm-operations/raci-matrix/raci-matrix.md` |
+| RAG Design Doc | `pm-ai` | `pm-ai/rag-design-doc/rag-design-doc.md` |
 | Red-Team Review | `pm-cross` | `pm-cross/red-team-review/red-team-review.md` |
 | Redundancy Consultation | `pm-hr` | `pm-hr/redundancy-consultation/redundancy-consultation.md` |
+| Referral Program Design | `pm-growth` | `pm-growth/referral-program-design/referral-program-design.md` |
 | Regex Builder & Explainer | `pm-engineering` | `pm-engineering/regex-builder/regex-builder.md` |
 | Renewal Playbook | `pm-cs` | `pm-cs/renewal-playbook/renewal-playbook.md` |
 | Research Protocol | `pm-research` | `pm-research/research-protocol/research-protocol.md` |
 | Retention Analysis | `pm-analytics` | `pm-analytics/retention-analysis/retention-analysis.md` |
+| Retention Loop Design | `pm-growth` | `pm-growth/retention-loop-design/retention-loop-design.md` |
 | Retrospective Analysis | `pm-delivery` | `pm-delivery/retro-analysis/retro-analysis.md` |
 | RFC Writer | `pm-engineering` | `pm-engineering/rfc-writer/rfc-writer.md` |
 | RICE + Strategic Alignment | `pm-planning` | `pm-planning/rice-impact-matrix/rice-impact-matrix.md` |

package/exports/aider/pm-ai/agent-spec/agent-spec.md

@@ -0,0 +1,61 @@
+# Agent Spec Skill
+
+An agent is a model plus tools plus a loop — and the danger lives in the tools and the loop, not the
+model. This skill specifies an agent so its *authority is explicit*: what it can do, what needs a human
+yes, and what happens when it's wrong. Scope and guardrails first; cleverness second.
+
+## Required Inputs
+
+Ask for these only if they aren't already provided:
+
+- **Job to be done** — the outcome the agent owns, and the boundary of its authority.
+- **Tools/actions** — what it can call (read APIs, write actions, code execution), and which are irreversible.
+- **Autonomy level** — fully autonomous, propose-then-approve, or co-pilot.
+- **Risk surface** — what's the worst thing a wrong action could do (spend money, send a message, delete data)?
+- **Success definition & escalation** — how "done" is judged, and when it must hand off to a human.
+
+## Output Format
+
+### Agent Spec: [name]
+
+**1. Goal & scope** — the job in one sentence; explicit non-goals and authority limits.
+
+**2. Tools / actions** — a table; mark each action's reversibility and required permission.
+
+| Tool | Purpose | Reversible? | Gate |
+|---|---|---|---|
+| search_kb | read context | yes | none |
+| send_email | notify | **no** | **human approval** |
+
+**3. Control loop** — plan → act → observe → reflect; the stopping condition; and a hard **max-steps / max-cost budget** so it can't loop forever.
+
+**4. Guardrails & approval gates** — which actions require a human yes (default: anything irreversible, outbound, or spending), input/output validation, and allow/deny lists. Pair irreversible actions with a dry-run preview (see [`action-runner`](../action-runner/SKILL.md)).
+
+**5. Memory & state** — what it remembers within a task vs. across tasks, and where (link a [`professional-brain`](../professional-brain/SKILL.md) for durable memory).
+
+**6. Escalation & handoff** — the triggers that stop the agent and route to a human (low confidence, repeated failure, out-of-scope request, high-risk action).
+
+**7. Evaluation** — task success rate, action correctness, and safety (false-action rate). Define with an [`ai-eval-plan`](../ai-eval-plan/SKILL.md), and test on adversarial/trap tasks.
+
+**8. Failure handling** — timeouts, tool errors, hallucinated tool calls, and the safe default (stop and ask, never guess on a high-risk action).
+
+## Quality Checks
+
+- [ ] Every tool is marked reversible/irreversible, and every irreversible action has a human gate
+- [ ] There is a hard max-steps and max-cost budget — the loop cannot run unbounded
+- [ ] Escalation triggers are explicit (confidence, repeated failure, out-of-scope, high-risk)
+- [ ] The safe default on uncertainty is "stop and ask", not "guess and act"
+- [ ] Evaluation includes a safety metric (wrong/unauthorised actions), not just task success
+- [ ] Non-goals and authority limits are stated, not implied
+
+## Anti-Patterns
+
+- [ ] Do not give an agent irreversible actions without an approval gate — autonomy and irreversibility together is how agents cause real damage
+- [ ] Do not omit a step/cost budget — an agent that can loop is an agent that can rack up cost or thrash forever
+- [ ] Do not measure only task success — an agent that completes the task by taking a wrong action has failed
+- [ ] Do not let the agent invent tool calls or arguments — validate against the schema and fail safe
+- [ ] Do not skip the "what's the worst case" analysis — the risk surface determines how many guardrails you need
+
+## Based On
+
+Tool-using / agentic design practice — bounded control loops, least-privilege tools, human-in-the-loop approval, and safety evaluation.

package/exports/aider/pm-ai/ai-eval-plan/ai-eval-plan.md

@@ -0,0 +1,59 @@
+# AI Eval Plan Skill
+
+You can't improve an AI feature you can't measure, and "it looks good in the demo" is not measurement.
+This skill produces an evaluation plan that turns a fuzzy quality goal into a repeatable, gated test —
+so a prompt change that quietly makes outputs worse can't ship.
+
+## Required Inputs
+
+Ask for these only if they aren't already provided:
+
+- **The feature & task** — what the model does and what "good output" means to a user.
+- **Failure modes that matter** — what bad looks like (hallucination, wrong format, unsafe, off-tone, too slow).
+- **Available data** — any real examples, logs, or labelled cases; or note there are none yet.
+- **Who judges quality** — automated checks, an LLM judge, human raters, or a mix.
+- **The decision this gates** — ship/no-ship, model selection, or prompt iteration.
+
+## Output Format
+
+### Eval Plan: [feature]
+
+**1. What we're measuring** — the task, and a one-line definition of a good vs. bad response.
+
+**2. Eval dataset**
+- **Cases:** how many, where they come from (real logs > synthetic), and how they're split (smoke set vs. full set).
+- **Coverage:** the slices/scenarios that must be represented (edge cases, adversarial, each major input type).
+- **Golden answers / references:** present or not, and how they were created.
+
+**3. Metrics & rubric**
+- **Per-dimension scores** — define each dimension (e.g. correctness, grounding, format, safety, tone) on an explicit 1–5 rubric with anchor descriptions, not vibes.
+- **Automated checks** — deterministic assertions first (valid JSON, contains required fields, no PII, latency budget).
+- **LLM-as-judge** — the judge prompt, the rubric it applies, and how you guard against its bias (calibrate against human labels on a sample).
+- **Human eval** — when it's required (safety, subjective quality) and the rater instructions.
+
+**4. Baselines** — what each candidate is compared against (current prompt, previous model, a plain-prompt control).
+
+**5. The bar** — the explicit threshold to ship (e.g. "≥4.2 avg correctness, 0 safety failures, p95 < 3s") and what happens if it's missed.
+
+**6. Regression gate** — how this runs in CI on every change, and the score-drop threshold that blocks a merge.
+
+## Quality Checks
+
+- [ ] Each metric has an explicit rubric with anchors — not just a name
+- [ ] Deterministic/automated checks are used wherever possible before reaching for an LLM judge
+- [ ] The LLM judge is calibrated against human labels on at least a sample
+- [ ] The eval set includes adversarial and edge cases, not just happy-path examples
+- [ ] There is a single, explicit numeric bar for the ship decision
+- [ ] The plan specifies how it runs as a regression gate, not just a one-time check
+
+## Anti-Patterns
+
+- [ ] Do not rely on a single overall score — a feature can pass on average while failing every safety case
+- [ ] Do not trust an LLM judge you haven't calibrated against humans — it has its own blind spots and biases
+- [ ] Do not eval only on happy-path inputs — the failures live in the edges and the adversarial cases
+- [ ] Do not let the eval set leak into the prompt/few-shot examples — that's training on the test set
+- [ ] Do not define the pass bar after seeing the scores — set the threshold before you run, or it means nothing
+
+## Based On
+
+LLM evaluation practice — task-grounded rubrics, LLM-as-judge with human calibration, and regression-gated CI evals.

package/exports/aider/pm-ai/ai-feature-prd/ai-feature-prd.md

@@ -0,0 +1,64 @@
+# AI Feature PRD Skill
+
+AI features break the normal PRD because the system is probabilistic: it will be wrong sometimes, and
+the product must be designed around that, not in denial of it. This skill extends a standard PRD with
+the AI-specific sections that decide whether the feature is trustworthy — the UX of uncertainty, the
+eval bar, guardrails, and what happens when the model is wrong.
+
+## Required Inputs
+
+Ask for these only if they aren't already provided:
+
+- **The user problem** and why an AI/probabilistic approach fits it (vs. deterministic rules).
+- **What "good" looks like** to the user, and the cost of a wrong answer (low-stakes vs. high-stakes).
+- **Inputs available** — context/data the model can use; privacy constraints.
+- **Trust level needed** — can the user verify the output, or must it be near-perfect?
+
+## Reads from / Writes to the Brain
+
+If a [`professional-brain`](../professional-brain/SKILL.md) exists, read `context.md` (product, users, voice)
+and `knowledge/strategy.md` first; write the feature to `entities/` and any scoping decision to `decisions/`,
+each provenance-tagged.
+
+## Output Format
+
+### AI Feature PRD: [feature]
+
+**1. Problem & why AI** — the user problem, and why a model (not rules) is the right tool. If rules would do, say so.
+
+**2. Experience** — the core flow, and crucially the **UX of uncertainty**: how confidence is shown, how the user verifies/edits, and how errors are made cheap to recover from. AI features live or die here.
+
+**3. Model approach** — prompt / fine-tune / RAG / agent (link [`rag-design-doc`](../rag-design-doc/SKILL.md) or [`agent-spec`](../agent-spec/SKILL.md)), the model tier, and why.
+
+**4. Quality bar & evaluation** — the metrics and the explicit ship threshold; reference an [`ai-eval-plan`](../ai-eval-plan/SKILL.md). State the acceptable error rate given the stakes.
+
+**5. Guardrails & safety** — what the feature must never do, input/output filtering, and handling of harmful/PII/out-of-scope inputs.
+
+**6. Fallback behaviour** — what happens when the model is unsure, wrong, slow, or down: graceful degradation, "I'm not sure" states, human handoff. **No silent confident errors.**
+
+**7. Data flywheel** — how usage (and the 👍/👎 / edits) feed back into evaluation and improvement, with the privacy boundary.
+
+**8. Cost & latency** — the per-request budget and p95 target; reference an [`llm-cost-latency-budget`](../llm-cost-latency-budget/SKILL.md).
+
+**9. Rollout** — staged exposure (internal → %→ GA), the guardrail metrics watched, and the rollback trigger.
+
+## Quality Checks
+
+- [ ] The PRD designs for the model being wrong — there's an explicit fallback, not just the happy path
+- [ ] The UX shows uncertainty and lets the user verify/correct cheaply
+- [ ] There's an explicit quality bar tied to the stakes (a medical answer and a tweet draft are not the same bar)
+- [ ] Guardrails name what the feature must never do
+- [ ] A data flywheel is defined with its privacy boundary
+- [ ] Cost and p95 latency budgets are stated, not left to "we'll see"
+
+## Anti-Patterns
+
+- [ ] Do not design only the happy path — a probabilistic feature without a fallback is a feature that fails loudly in production
+- [ ] Do not hide uncertainty behind a confident UI — overclaimed confidence is how AI features lose user trust permanently
+- [ ] Do not use AI where deterministic rules are better, cheaper, and more reliable — "AI" is not the goal
+- [ ] Do not set one quality bar for all stakes — calibrate the acceptable error rate to the cost of being wrong
+- [ ] Do not ship without a rollback trigger and guardrail metrics — a probabilistic system needs a kill switch
+
+## Based On
+
+Standard PRD practice (see [`prd-template`](../prd-template/SKILL.md)) extended for probabilistic systems — uncertainty UX, eval gates, guardrails, and graceful fallback.