← Back to Skills Marketplace
cpsean

Product Review Panel Skill

by Sean Liu · GitHub ↗ · v1.0.0 · MIT-0
cross-platform ✓ Security Clean
35
Downloads
0
Stars
1
Active Installs
1
Versions
Install in OpenClaw
/install product-review-panel-skill
Description
Convene a multi-expert panel to review a Product Requirements Document (PRD) and deliver a binding verdict (GO / NO-GO / CONDITIONAL GO) with dissenting opin...
README (SKILL.md)

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:

  1. 倾向标签: 倾向 GO / 倾向 NO-GO / 倾向 CONDITIONAL
  2. ≤ 80-word rationale in their voice and framework
  3. 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.

  1. 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.

  2. 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
  3. Maximum 5 intake rounds. If the PM is uncooperative, proceed with assumptions. Do NOT loop forever asking for info.

  4. No numeric scoring. Tendency labels only. Numbers create false precision.

  5. 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."

  6. 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 output
  • example-intl-pricing-review.md — English pricing change PRD → full review output

Refer to these when calibrating tone, length, and structure.

Usage Guidance
Install this if you want a structured, opinionated PRD review panel. Be aware that Chinese conversations automatically use the China-oriented panel; if that is not the perspective you want, explicitly ask for the international panel or a different review frame. Avoid sharing sensitive PRDs or private document links unless you are comfortable with the runtime reading them for the review.
Capability Assessment
Purpose & Capability
The artifacts consistently support the stated purpose: convening a structured multi-expert panel to review PRDs and product proposals, then issuing GO / NO-GO / CONDITIONAL GO guidance with dissent preserved.
Instruction Scope
The Chinese-versus-English panel selection is explicit in the description and SKILL.md, and the scanner concern is a real framing caveat, but it affects review perspective rather than granting hidden authority or access.
Install Mechanism
The package is markdown-only, declares no dependencies, has no executable scripts, and no install hooks or package commands were found.
Credentials
The skill may read user-provided PRDs, uploaded files, or connected document links via runtime-supported tools, which fits the review purpose but means users should be intentional about which business documents they provide.
Persistence & Privilege
No background persistence, privilege escalation, credential/session access, local indexing, automatic mutation, or external transfer mechanism was found.
How to Use
  1. Make sure OpenClaw is installed (local or Docker)
  2. Run the install command in chat: /install product-review-panel-skill
  3. After installation, invoke the skill by name or use /product-review-panel-skill
  4. Provide required inputs per the skill's parameter spec and get structured output
Version History
v1.0.0
Initial release of the product-review-panel skill: - Launches a structured, multi-expert panel to review Product Requirements Documents (PRDs) and deliver a clear verdict (GO / NO-GO / CONDITIONAL GO), with dissenting opinions always preserved. - Dynamically adapts expert roster based on user conversation language (Chinese or English) and PRD classification. - Enforces a rigorous review flow: intake interrogation, expert opinions, pointed debate, authoritative verdict by "The Closer," and explicit inclusion of dissent. - Applies strict behavioral guardrails: no numeric scoring, no personal attacks, mandatory disclaimer, one verdict per run. - Accepts flexible PRD input formats (uploaded files, pasted text, document links). - Not for brainstorming, technical architecture, or non-product questions—use other skills for those cases.
Metadata
Slug product-review-panel-skill
Version 1.0.0
License MIT-0
All-time Installs 1
Active Installs 1
Total Versions 1
Frequently Asked Questions

What is Product Review Panel Skill?

Convene a multi-expert panel to review a Product Requirements Document (PRD) and deliver a binding verdict (GO / NO-GO / CONDITIONAL GO) with dissenting opin... It is an AI Agent Skill for Claude Code / OpenClaw, with 35 downloads so far.

How do I install Product Review Panel Skill?

Run "/install product-review-panel-skill" in the OpenClaw or Claude Code chat to install it in one step — no extra setup required.

Is Product Review Panel Skill free?

Yes, Product Review Panel Skill is completely free, licensed under MIT-0. You can download, install and use it at no cost.

Which platforms does Product Review Panel Skill support?

Product Review Panel Skill is cross-platform and runs anywhere OpenClaw / Claude Code is available (cross-platform).

Who created Product Review Panel Skill?

It is built and maintained by Sean Liu (@cpsean); the current version is v1.0.0.

💬 Comments