---
name: prd-stress-test
description: Run seven stakeholder personas against a PRD in parallel and get a ranked list of what to fix. Use this skill when: PRD ready for review — Fired manually or on a status change in the doc.
---

# Stress-test the spec before it ships

Goal: stress-test a PRD from 7 stakeholder personas and produce a ranked fix list.

Connected tools this skill uses: drive, notion, slack.

## When to use

- PRD ready for review — Fired manually or on a status change in the doc.
- Or when the user asks directly to stress-test a PRD from 7 stakeholder personas and produce a ranked fix list.

## Method

Run the 7-persona PRD stress test.

1. Read the PRD at {{prd_path}} and confirm it has a Problem, Non-goals, and
   Requirements section; if any is missing, stop and flag the gap instead of
   scoring.
2. Fan out to 7 independent sub-agents, one per persona: eng lead, designer,
   CFO, sales, support, legal, and a skeptical exec. Run all 7 concurrently
   in the cloud, not in sequence.
3. Each sub-agent scores the PRD 0-100 across 5 dimensions —
   {{scoring_dimensions}} — from its persona's stake, and returns a
   ship / revise / reject verdict with its reasoning.
4. Validation: confirm all 7 sub-agents returned a full score set (all 5
   dimensions) and a verdict; re-run any persona that came back incomplete
   before synthesizing.
5. Synthesize the 7 reviews into one prioritized fix list: blockers first,
   then the disagreements between personas worth flagging on their own.
6. Post the scorecard (7 verdicts, top 5 fixes) to {{delivery_channel}} and
   save the full teardown to
   `reports/prd-stress-test/{{prd_slug}}-{{date}}.md`.

Never edit the PRD directly — every fix is a recommendation for the PRD's
owner to accept or reject.

## Persona scoring rubric

Score a PRD as one stakeholder persona, against fixed dimensions and anchors, not gut feel.

## Method
1. Read the PRD fully from your persona's perspective — eng lead, designer, CFO, sales, support, legal, or skeptical exec.
2. Score each of the 5 dimensions — {{scoring_dimensions}} (set on first run, then stored in memory) — from 0 to 100.
3. Anchor each score: 30 means the dimension is barely addressed or actively wrong for this persona's stake; 70 means it's workable with named gaps; 90 means it fully holds up to this persona's scrutiny.
4. Write one sentence of reasoning per dimension score — no unexplained numbers.
5. Roll the 5 scores into one verdict: ship, revise, or reject, using the team's stored threshold (set on first run, then stored in memory).
6. List the specific gaps that pulled the score down, in order of severity.

## Rules
- Score only from your assigned persona's stake — an eng lead does not score business case the way a CFO would.
- A score below the reject threshold on any single dimension (set on first run, then stored in memory) forces a reject verdict regardless of the average.
- Every score needs a one-line reason; an unexplained score is incomplete and gets re-run.
- Stay consistent with the anchors across runs — do not drift the meaning of a 70 between PRDs.

## Output
Five dimension scores (0-100) with one-line reasoning each, one verdict (ship/revise/reject), and a ranked list of gaps.

## Cross-persona synthesis

Merge 7 independent persona reviews into one prioritized fix list, without double-counting overlapping gaps.

## Method
1. Collect all 7 persona scorecards, each with its 5 dimension scores, verdict, and gap list.
2. Pool every gap across all 7 reviews into one list.
3. Merge gaps that describe the same underlying issue from different personas — note which personas raised it, don't list it twice.
4. Order the merged list blockers first: anything tied to a reject verdict from any persona outranks anything tied only to a revise.
5. Flag genuine disagreements separately — where personas reached opposite verdicts or conflicting recommendations on the same point — instead of averaging them into a single score.
6. Confirm every one of the 7 verdicts is represented somewhere in the final output before closing it out.

## Rules
- Never average away a disagreement between personas — a legal reject and a sales ship on the same clause is a flag, not a wash.
- Blockers (any reject-tied gap) always rank above revise-tied gaps, regardless of which persona raised them.
- Do not drop a persona's finding just because another persona didn't raise it — a single-persona blocker still blocks.
- Keep the merged list traceable: every fix item names which persona(s) raised it.

## Output
One prioritized fix list (blockers first), a separate short list of flagged persona disagreements, and the 7 source verdicts for reference.

## PRD completeness check

Confirm the PRD has Problem, Non-goals, and Requirements sections filled in before any persona scores it.

## Method
1. Open the PRD at {{prd_path}} and locate the Problem, Non-goals, Requirements, and metrics sections.
2. Check that each required section exists and has real content — a heading with no text under it does not count.
3. Check that the Non-goals section lists actual exclusions, not a placeholder line like "TBD."
4. Check that any quantitative claim in the PRD has a source or is marked "[baseline needed: ...]" rather than left bare.
5. If every required section passes, clear the PRD to move to persona scoring.
6. If any section is missing or empty, stop before fan-out and return the specific gap instead of scoring an incomplete draft.

## Rules
- Non-goals is mandatory — a PRD with an empty or missing Non-goals section does not proceed to scoring.
- Do not guess at missing content to let the PRD pass; flag the gap and stop.
- This check runs once per PRD version — re-run it if the PRD is revised and resubmitted.
- Completeness is a gate, not a score — it blocks fan-out, it does not count as one of the 5 scored dimensions.

## Output
A pass/fail result. On fail: the specific missing or empty section(s), returned instead of a scorecard. On pass: clearance to fan out.