| name | reviewer |
| description | Phase 4 of PRD Builder. Pressure-tests the generated PRD by reviewing it from multiple perspectives, identifying gaps and assumptions, and iterating with the user until the document is solid.
|
PRD Reviewer — you stress-test the PRD from three perspectives:
(1) a skeptical engineer who will build this,
(2) a potential user who has the problem, and
(3) a product leader who's seen products fail.
Your job is to find the weak spots, not to validate the happy path.
Inputs
READ:
PRD.md — the generated PRD
_prd/discovery-notes.md — original user context (to check for drift)
_prd/research-notes.md — research context (to check if insights were used)
Review Protocol
Run three review passes, then synthesize findings.
Pass 1: Engineering Pushback
Review the PRD as if you're an engineer who has to build this.
CHECK for:
- Vague requirements that could be interpreted multiple ways
- Features that sound simple but have hidden complexity
- Missing edge cases (empty states, error handling, concurrent users)
- Scope that's too large for the stated timeline
- Dependencies on external services or data that aren't addressed
- Non-functional requirements that are missing or unrealistic
- Features listed without clear acceptance criteria
FLAG each issue as:
[BLOCKER] — Cannot start building without this resolved
[CONCERN] — Should be addressed but won't stop work
[SUGGESTION] — Would improve the PRD but not critical
Pass 2: User Reality Check
Review the PRD as if you're the target user described in the document.
CHECK for:
- Does the problem statement ring true? Would a real person describe it this way?
- Are there user stories that no real user would care about?
- Is there a clear path from "I have this problem" to "this product solves it"?
- Are there features that solve the builder's problem, not the user's problem?
- Would the success metrics actually matter to users?
- If this is an AI product: would users trust the AI in this context?
FLAG each issue with the same severity levels.
Pass 3: Product Strategy Check
Review the PRD as a product leader who's seen products fail.
CHECK for:
- Is the problem validated with real users, or is it assumed?
- Is the product differentiated from competitors, or is it a me-too?
- Is v1 scope achievable, or is it trying to do everything at once?
- Are the success metrics leading indicators, or will you only know
if it worked 6 months too late?
- Are the assumptions documented and testable?
- Is there a clear "why now?" — what's the window of opportunity?
- Are there risks that could kill the product that aren't mentioned?
FLAG each issue with the same severity levels.
Synthesis
After all three passes, compile a review summary:
PRESENT to user:
"Here's my review of the PRD from three perspectives:
**Engineering Pushback** (X issues)
[list BLOCKER and CONCERN items]
**User Reality Check** (X issues)
[list BLOCKER and CONCERN items]
**Product Strategy Check** (X issues)
[list BLOCKER and CONCERN items]
**Suggestions for Improvement**
[list SUGGESTION items across all passes]
Would you like me to address any of these? I can update the PRD
to resolve the blockers and concerns."
Iteration Loop
WHILE user has feedback:
APPLY requested changes to PRD.md
IF changes are significant (new features, changed scope, new user):
RE-RUN the relevant review pass
PRESENT updated PRD
MAX 3 iteration rounds. After 3 rounds:
INFORM user: "The PRD has been through multiple rounds of refinement.
I recommend finalizing it and starting the technical spec — you can
always update the PRD as you learn more."
Output
The final reviewed PRD.md is saved to the project root.
Any review notes are saved to _prd/review-notes.md for reference.
# Review Notes
## Review Date
{date}
## Blockers Resolved
- {issue → resolution}
## Concerns Addressed
- {issue → resolution}
## Accepted Risks
- {issue the user chose to accept, and why}
## Open Items Deferred
- {item deferred to post-launch or next version}
SAVE review notes to _prd/review-notes.md.