AGENT SKILL

mh-cto-sdlc-prd-validate

暂无简介

适用职业产品经理
License 许可
MIT
最近更新
2026-05-13
Stars
★ 2
Forks
0

02 资源文件 2

03 SKILL.md 正文

PRD Validate

Validation is adversarial. Your job is not to confirm the PRD looks good — it is to find what the PRD got wrong, invented without evidence, or failed to cover. Every finding must be specific: name the PRD section, name the Phase 1 artifact that contradicts or doesn't support it.

Step 1 — Load

Confirm both artifacts are present:

  • PRD (_mh-output/cto/sdlc/prd-create/...prd.md) ✓ / ✗
  • All Phase 1 artifacts via persistent_facts ✓ / ✗

If PRD is missing: "Cannot validate. Run /mh-cto-sdlc-prd-create first."

Step 2 — Traceability Audit

For every Must-priority feature in the PRD Feature Catalogue:

Check 1 — Source exists Does the feature have a named source in Phase 1 (journey map pain point, research finding, or problem statement)? If the PRD lists a feature with no source: INVENTED — flag it.

Check 2 — Source supports priority Is the named source actually high-pain (pain level 2-3 in journey map, or Tier 1 research finding)? A feature sourced from a minor observation should not be Must-priority. Downgrade recommendation → PRIORITY MISMATCH.

Check 3 — Domain model coverage Is every feature's subject entity present in the domain model? If the feature references an entity not in the domain model: DOMAIN GAP — the domain model needs updating before epic mapping.

Step 3 — Goal Coverage Audit

For every Goal in the PRD:

Check 4 — Feature coverage Does at least one Must-priority feature directly serve this goal? A goal with no Must feature is aspirational noise — flag it.

Check 5 — Metric exists Does each goal have a named success metric? A goal without a metric cannot be validated at launch.

Check 6 — Persona coverage Does the goal serve at least one named persona? A goal that serves no one in the stakeholder map should not be in the PRD.

Step 4 — Non-Goals Audit

Check 7 — Hard constraints reflected Does every hard constraint from the constraint inventory appear in Non-Goals or Constraints Summary? If a hard constraint is not represented anywhere in the PRD: CONSTRAINT BLINDSPOT — the PRD is committing to something the constraint prohibits.

Check 8 — Journey map out-of-scope reflected Does the PRD's Non-Goals list match the journey map's out-of-scope definition? Things the journey map excluded that the PRD includes are scope creep.

Step 5 — Assumption Audit

Check 9 — Active Unverified assumptions visible Does the PRD's Active Assumptions section include all Unverified assumptions from the assumption list? Missing unverified assumptions means the PRD is hiding its own risk.

Check 10 — No Invalidated assumptions as foundation Does any PRD feature rest on an assumption that Phase 1 invalidated? This is a phase 0 re-entry flag passed forward.

Step 6 — Conditional Pass Requirements (if applicable)

If the Phase 1 gate verdict was Conditional Pass, check each named Phase 2 requirement from the gate verdict:

Check 11 — Every Conditional Pass requirement addressed Read the gate verdict's Phase 2 requirements. Is each one addressed in the PRD? A PRD that ignores a gate requirement is not valid.

Step 7 — Build Validation Report

After all checks, compile findings:

Finding tiers:

  • BLOCKING — Must fix before epic mapping begins. (Invented Must features, constraint blindspots, missing gate requirements)
  • REQUIRED — Must fix before story-spec begins. (Priority mismatches, goal without metric, missing assumption)
  • ADVISORY — Improve now or note it. (Could-priority features without sources, minor wording mismatches)

For each BLOCKING or REQUIRED finding:

  • Name the finding type (from check list above)
  • Name the specific PRD section and line
  • Name the Phase 1 artifact that surfaces the conflict
  • State what must change

Step 8 — Resolve Findings

For each BLOCKING finding: ask the user to resolve it now. One at a time.

For each REQUIRED finding: present all at once and ask the user to address them before moving to story-spec.

For ADVISORY findings: list them and ask if the user wants to address them now or log them.

After all BLOCKING and REQUIRED findings are resolved, update the PRD with the user's resolutions and save the updated version.

Step 9 — Write Validation Report

Save to _mh-output/cto/sdlc/prd-validate/{YYYY-MM-DD}_{r###}_prd-validation.md:

# PRD Validation Report

**Date:** {date}
**PRD Version:** {from PRD header}
**Result:** Clean | Requires Revision

## Findings

### BLOCKING

{list or "None"}

### REQUIRED

{list or "None"}

### ADVISORY

{list or "None"}

## Resolution Log

| Finding   | Resolution     | Status              |
| --------- | -------------- | ------------------- |
| {finding} | {what changed} | Resolved / Deferred |

## Validation Result

{Clean: "PRD is traceable and consistent. Proceed to epic mapping."}
{Requires Revision: "PRD has [N] unresolved BLOCKING findings. Do not proceed."}

Step 10 — Present Summary

PRD Validation: {Clean | Requires Revision}

Checks passed: {N}/{11}

Findings: {N} BLOCKING, {N} REQUIRED, {N} ADVISORY

Next step: {If clean → Run /mh-cto-sdlc-epic-map} {If not → list what must be fixed}

04 同仓库更多 Skills

Imad-Oute
Imad-Oute/MicroHard

mh-vo-vault-brief

Weekly vault intelligence report — health score, pipeline status, OS alignment, and top issues. Use when the user says vault brief, weekly vault report, vault intelligence, run the vault brief, or weekly review brief.

★ 2
Imad-Oute
Imad-Oute/MicroHard

mh-vo-output-bridge

Scan _mh-output/ for new artifacts and create or update corresponding vault entries. Use when the user says run the bridge, sync outputs to vault, output bridge, create vault entries, sync mh-output to vault, or bridge outputs.

★ 2
Imad-Oute
Imad-Oute/MicroHard

mh-vo-moc-builder

Create or update Maps of Content for vault sections and ensure every note is reachable from a MOC. Use when the user says build MOCs, update MOCs, rebuild maps of content, MOC builder, create maps of content, or update the MOCs.

★ 2
Imad-Oute
Imad-Oute/MicroHard

mh-vo-link-architect

Identify missing vault connections and generate precise link recommendations. Use when the user says recommend links, find missing connections, link architect, what should be connected, or identify missing links.

★ 2
Imad-Oute
Imad-Oute/MicroHard

mh-vo-base-builder

Generate Obsidian Bases (.base files) for live database views of vault data. Use when the user says build base views, create base files, generate bases, build live views, or create database views.

★ 2
Imad-Oute
Imad-Oute/MicroHard

mh-vo-taxonomy-enforcer

Audit vault tags and frontmatter, find drift, and apply corrections. Use when the user says fix the taxonomy, enforce tags, fix frontmatter, clean up tags, taxonomy enforcer, or fix tags.

★ 2