| Hat |
Stage |
Upstream gate |
Output |
Blocks |
| PO |
Product review |
PRD at draft + epic approved |
product/outputs/prds/PRD-NNN-review.md |
wb.approve PRD-NNN on any P0 |
When to Use
- PRD is at
status: draft and author wants gap-finding before publishing.
- Author about to
wb.publish PRD-NNN and wants panel sign-off first.
Prerequisites
- PRD file at
product/outputs/prds/PRD-{NNN}-*.md with status: draft.
- Epic-context approved in
.workbench-state/approved.json (unapproved epic → waste; refuse).
Reviewers (Dispatched in Parallel)
| # |
Reviewer |
Lens |
| 1 |
Engineering |
Feasibility, dependencies, scalability, edge cases, estimate realism |
| 2 |
Design |
UX, interaction patterns, IA, missing states |
| 3 |
Executive |
Strategic fit, opportunity cost, ROI, portfolio effects |
| 4 |
Legal / Compliance |
Data handling, regulatory exposure, contract impact, a11y law |
| 5 |
UX Research |
Validation evidence, unvalidated assumptions, research gaps |
| 6 |
Skeptic |
Challenge every assumption, unstated risks, attack the hypothesis |
| 7 |
Customer Voice |
First-person persona reaction |
Every reviewer outputs: ✅ What works / ⚠️ Concerns (P0/P1/P2) / ❌ Blockers / 💡 Suggestions. Must cite PRD section numbers (e.g. §4 AC-2).
Protocol
- Read PRD in full (problem, hypothesis, goal, stakeholders, ACs, non-goals, metrics, rollout, approach, risks).
- Resolve
stage from frontmatter (kickoff | planning | xfn | solution | launch); default planning.
- Dispatch 7 reviewers in parallel (single message, 7 Task calls) — each gets full PRD + stage + context-library.
- Synthesise to
product/outputs/prds/PRD-{NNN}-review.md with Consensus / Disagreements / Priority fix list (🔴 P0 / 🟡 P1 / 🟢 P2).
- Count P0. If > 0, block and instruct author to revise + re-run.
Flags
--perspectives "eng,design,skeptic" — subset mode.
--stage {name} — override frontmatter stage.
Do Not
- Collapse verbatim reviewer output — verbatim goes into the synthesis.
- Auto-approve, even with zero P0s. User calls
wb.approve.
- Dispatch fewer than requested reviewers — run empty ones (“no material legal exposure”) explicitly.
- Skip section-number citations.
Source
skills/prd-review-panel/SKILL.md