Product Review Panel
Convene a multi-expert panel to review a Product Requirements Document (PRD) and deliver a binding verdict (GO / NO-GO / CONDITIONAL GO) with dissenting opinions preserved as first…
Sean Liu
@cpsean
Install
$ openclaw skills install @cpsean/product-review-panel-skill产品评审团 / Product Review Panel
This skill convenes a multi-expert panel to review a PRD or product proposal, runs a structured discussion, and delivers a definitive verdict with preserved dissent.
When to invoke
User asks for any of:
- "评审 PRD" / "review my PRD" / "panel review"
- "这个功能要不要做" / "should I build this feature"
- Pastes a PRD and asks for critical feedback
- Asks for a "second opinion" / "expert review" on a product proposal
Language detection (determines panel)
Use the user's conversation language, NOT the PRD's content language:
- Conversation in 中文 → load
references/personas/experts-cn.md(core: Marty Cagan + 俞军 + 大厂 P9 产品总监) - Conversation in English or other → load
references/personas/experts-intl.md(core: Marty Cagan + Clayton Christensen + Senior PM Director)
If a Chinese-speaking PM asks to review an English PRD, still use the Chinese panel — the PM is the audience.
Execution flow
Run these steps in order. Each step references its detailed instructions file.
Step 0 — Disclaimer
Print the disclaimer from references/templates/disclaimer.md at the very top. Always show it, every run. Non-negotiable.
Step 1 — Information gap check (P9 / Senior PM Director intake)
Load references/personas/p9-director.md for the character definition and references/templates/intake-dialogue.md for the dialogue format.
The P9 / Senior PM Director conducts a pre-review interrogation:
- Read the PRD
- Identify gaps using the checklist in
references/workflows/information-gap-check.md - Ask at most 5 turns of 1-2 focused questions each
- If PM answers: log answer, move on
- If PM skips: brief in-character snark, log the skip, continue (do NOT loop)
- If PRD is already comprehensive: one-line acknowledgment, no questions, proceed
Track all skipped items — they become evidence for The Closer in Step 7.
When intake ends, P9 delivers the panel announcement (see references/templates/intake-dialogue.md — closing sequence): states the PRD type, names each situational expert and what they will focus on, and gives the total panel count. This announcement is in P9's voice and precedes the formal panel intro card.
Step 2 — PRD classification & panel composition
Load references/workflows/prd-classification.md. Classify the PRD into one of:
- 新功能 (new feature)
- 迭代优化 (iteration)
- 商业模式 / 定价 (business model / pricing)
- 体验重构 (UX redesign)
- 早期探索 (early-stage exploration)
Based on the classification, select 1-2 situational experts and combine with the 3 core experts for the active language.
Step 3 — Panel intro card
Print the "出场卡" using format from references/templates/panel-intro-card.md. One row per expert: name + credential + framework + signature question.
Step 4 — Round 1: parallel reviews
Each expert (loaded from references/personas/experts-{cn|intl}.md) gives exactly:
- 倾向标签: 倾向 GO / 倾向 NO-GO / 倾向 CONDITIONAL
- ≤ 80-word rationale in their voice and framework
- One follow-up question they'd want the PM to answer
No scoring numbers. Only tendency labels. Numbers create false precision.
Step 5 — Tendency direction check
- If GO and NO-GO both appear in tendencies → proceed to Step 6
- If unanimous (all GO or all NO-GO or all CONDITIONAL) → skip Step 6, but The Closer must challenge the consensus in Step 7
Step 6 — Round 2: pointed debate
Pick the strongest GO-leaning expert and the strongest NO-GO-leaning expert. One exchange only:
- GO expert states the strongest case against the NO-GO expert's position
- NO-GO expert responds
- End. No further rounds. No relay debate.
Step 7 — The Closer: final verdict
Load references/personas/closer.md and references/workflows/verdict-logic.md.
The Closer:
- Tallies tendencies
- Runs the verdict decision tree
- Quotes 1-2 specific expert phrases as supporting evidence
- Cites P9's skip log if applicable
- Issues final verdict
- If CONDITIONAL GO: lists concrete conditions, each with a deadline
- Lists 2+ "翻车前兆信号" — observable signals that would invalidate the verdict
- Closes with "完。" (Chinese) / "Done." (English)
Step 8 — Dissent section (Dissent)
Print the dissent block using format from references/templates/output-structure.md:
- Which expert(s) held the dissenting view
- Their strongest argument verbatim
- "未来翻车前兆信号" / "Future failure signals" — what to monitor
Dissent is always included, even if minor. It is a first-class output.
Global guardrails
These are hard constraints. The skill must never violate them.
-
All expert "perspectives" are interpretive applications of public frameworks, not actual statements by the real individuals. P9 / Senior PM Director / The Closer are explicit fictional archetypes. The Step 0 disclaimer makes this explicit to the user.
-
Persona banned behaviors:
- No personal attacks on the PM ("你不懂吗" / "you don't get it" / "are you stupid")
- No lecturing beyond 2 sentences in a single turn
- No profanity or aggressive insults
- Critique the PRD, not the person — full list in each persona file
-
Maximum 5 intake rounds. If the PM is uncooperative, proceed with assumptions. Do NOT loop forever asking for info.
-
No numeric scoring. Tendency labels only. Numbers create false precision.
-
The Closer must deliver a verdict every run. If genuinely insufficient information, the verdict is CONDITIONAL GO with explicit conditions, not "we need more data."
-
Dissent must always be preserved. Even when verdict is unanimous, if a minor concern existed in Round 1, it goes in the dissent section.
Input handling
Accept any input form the runtime supports:
- Uploaded files (
.md,.docx,.pdf, etc.) — use Read tool - Pasted text in the conversation
- Links to documents in connected systems (Notion, Confluence, etc.) — use the relevant MCP tool
If no PRD is provided at invocation, ask the PM to share one before starting Step 0.
File reference map
| Step | Reads |
|---|---|
| 0 | references/templates/disclaimer.md |
| 1 | references/personas/p9-director.md, references/templates/intake-dialogue.md, references/workflows/information-gap-check.md |
| 2 | references/workflows/prd-classification.md, `references/personas/experts-{cn |
| 3 | references/templates/panel-intro-card.md |
| 4 | `references/personas/experts-{cn |
| 6 | `references/personas/experts-{cn |
| 7 | references/personas/closer.md, references/workflows/verdict-logic.md |
| 8 | references/templates/output-structure.md |
Examples
End-to-end worked examples in references/examples/:
example-cn-feature-review.md— Chinese new feature PRD → full review outputexample-intl-pricing-review.md— English pricing change PRD → full review output
Refer to these when calibrating tone, length, and structure.
Top skills in this category
Playwright (Automation + MCP + Scraper)
@ivangdavilaAutomates, tests, and debugs browsers with Playwright: locators, auto-waiting, traces, CI runs, and MCP browser control. Use when a test is flaky, times out, or fails only in CI or headless; when a locator matches multiple elements or the wrong one (strict mode violation); when clicks need force, waits become sleeps, or networkidle never settles; for storageState and login setup, request mocking and HAR replay, uploads and downloads, iframes and shadow DOM, popups and dialogs, screenshot diffs that change per machine, trace and report artifacts, sharding a slow suite, device and permission emulation, accessibility checks, driving a real browser through Playwright MCP, extracting data from JS-rendered pages, or porting a Cypress, Puppeteer, or Selenium suite to Playwright. Not for maintaining an existing Cypress or Puppeteer suite (cypress, puppeteer) or for work a plain HTTP request answers (http).
Openclaw Command Center
@jontsaiMission control dashboard for OpenClaw - real-time session monitoring, LLM usage tracking, cost intelligence, and system vitals. View all your AI agents in o...
Computer Use
@ram-raghav-sFull desktop computer use for headless Linux servers. Xvfb + XFCE virtual desktop with xdotool automation. 17 actions (click, type, scroll, screenshot, drag,...
Interview Simulator
@wscatsSimulates mock interviews for any role and experience level with tailored technical, behavioral, and case questions plus detailed feedback and scoring.
Scrapling Official Skill
@d4vinciScrape web pages using Scrapling with anti-bot bypass (like Cloudflare Turnstile), stealth headless browsing, spiders framework, adaptive scraping, and JavaScript rendering. Use when asked to scrape, crawl, or extract data from websites; web_fetch fails; the site has anti-bot protections; write Pyth