AgentSkillsCN

prd-validate

针对 PRD 进行轻量级的语法检查与一致性校验（速度远超完整分析）。

SKILL.md
--- frontmatter
name: prd-validate
description: Lightweight lint and consistency check for a PRD (faster than full analysis)
argument-hint: <path-to-prd.md>
user-invocable: true
disable-model-invocation: true
allowed-tools: [Read, Glob, Grep]

User Input

text
$ARGUMENTS

You MUST consider the user input before proceeding (if not empty).

Goal

Perform a fast, mechanical lint check on a PRD file. Unlike /prd-analyze (which does deep semantic analysis), this checks structural and formatting issues only. Output is a pass/fail report displayed in the terminal — no files are modified.

Execution Steps

1. Resolve PRD Path & Detect Type

  • •If $ARGUMENTS is empty, abort with: "Please provide a path to a PRD file. Example: /prd-validate prd/my-feature/prd.md"
  • •Read the file. If it does not exist, abort with a clear error message.
  • •Detect PRD type by scanning the first 50 lines for a Type: or **Type:** metadata field using Grep:
    • •Pattern: Type:\s*(Full PRD|1-Pager|Shape Up Pitch|AI Product PRD)
    • •Match against these exact values:
      • •"Full PRD" → Full PRD template (22 sections, 54-point rubric)
      • •"1-Pager" → 1-Pager template (7 sections, 20-point rubric)
      • •"Shape Up Pitch" → Shape Up Pitch template (6 sections, 16-point rubric)
      • •"AI Product PRD" → AI Product PRD template (27 sections, 64-point rubric)
    • •If no match found: Default to "Full PRD" and add a WARNING finding: "No Type metadata found, defaulting to Full PRD validation"
  • •Store the detected type for use in subsequent checks

2. Run Lint Checks

Perform these mechanical checks (no semantic analysis):

A. Section Structure

  • • Has a top-level # heading (document title)
  • • All required section headings are present based on detected type (fuzzy match against template names):
    • •Full PRD: 11 required of 22 total sections (Metadata, Executive Summary, Problem Statement, Goals/KPIs, Target Users, User Stories, Functional Requirements, Data Model, Error States, Non-Functional Requirements, Acceptance Criteria)
    • •1-Pager: 7 required of 7 total (ALL sections: Metadata, Problem, Hypothesis, Goals, Solution Overview, Key Risks, Timeline)
    • •Shape Up Pitch: 6 required of 6 total (ALL sections: Metadata, Problem, Appetite, Solution, Rabbit Holes, No-Gos)
    • •AI Product PRD: 18 required of 27 total (all Full PRD required + 5 AI-specific: Model Performance Requirements, Data Dependencies & Pipeline, Failure Modes & Fallbacks, Ethical Considerations & Bias, Monitoring & Drift Detection)
  • • Section headings use ## level (not ### or deeper for top-level sections)
  • • Sections appear in template order (note any out-of-order sections, but this is non-critical)

B. Metadata Completeness

  • • Has Version field
  • • Has Status field (one of: Draft, In Review, Approved, Active Development)
  • • Has Last Updated date
  • • Has Author field
  • • Has a Revision History table

C. Placeholder Detection

  • • No [NEEDS INPUT] markers
  • • No TODO or TBD markers
  • • No ??? or TKTK markers
  • • No <placeholder> style markers
  • • No "to be determined" / "to be decided" phrases

D. Requirement Formatting (Conditional based on type)

  • • Functional requirements have IDs (e.g., FR-XXX-##) — Full PRD and AI PRD only
  • • Requirements have priority labels (P0/P1/P2 or Must/Should/Could) — Full PRD and AI PRD only
  • • Acceptance criteria use Given/When/Then format (or equivalent structured format) — Full PRD and AI PRD only
  • •Note: For 1-Pager and Shape Up Pitch, these checks are marked as N/A (automatically PASS) since those templates use different formats

E. Table Integrity

  • • All markdown tables have header rows
  • • Tables are not broken (consistent column counts)
  • • KPI table exists with numeric targets

F. Content Indicators

  • • Document is not a stub (> 200 lines)
  • • No empty sections (heading followed immediately by another heading)
  • • Enumeration values are explicit (no "e.g." or "such as" in enum definitions)

3. Produce Validation Report

Output a Markdown report with type-aware formatting:

code
## PRD Validation Report

**File**: <path>
**Detected Type**: <Full PRD | 1-Pager | Shape Up Pitch | AI Product PRD>
**Template**: <prd-template.md | prd-template-1pager.md | prd-template-pitch.md | prd-template-ai.md>
**Expected Sections**: <22 | 7 | 6 | 27> (<required count> required)
**Date**: <current date>
**Result**: PASS / FAIL (X issues)

---

### Checks

| # | Check | Result | Details |
|---|-------|--------|---------|
| 1 | Document title present | PASS | — |
| 2 | Required sections present | FAIL | Missing: Acceptance Criteria, Error States |
| 3 | Metadata: Version | PASS | v1.0 |
| 4 | Metadata: Status | PASS | Draft |
| 5 | Metadata: Date | PASS | 2026-01-31 |
| 6 | Metadata: Author | FAIL | Not found |
| 7 | Revision history | PASS | — |
| 8 | No placeholders | FAIL | 3 found: L42, L88, L156 |
| 9 | Requirement IDs | PASS / N/A | (N/A for 1-Pager, Pitch) |
| 10 | Priority labels | PASS / N/A | (N/A for 1-Pager, Pitch) |
| 11 | Acceptance criteria format | PASS / N/A | (N/A for 1-Pager, Pitch) |
| 12 | Table integrity | PASS | — |
| 13 | KPI table with targets | PASS | — |
| 14 | Not a stub (> 200 lines) | PASS | 1,329 lines |
| 15 | No empty sections | PASS | — |
| 16 | Enum completeness | PASS | — |

---

### Summary

- **Passed**: X / 16 (including N/A checks)
- **Failed**: X / 16
- **N/A**: X / 16 (checks not applicable to this template type)
- **Result**: PASS — PRD structure is valid / FAIL — X issues need attention

### Suggested Actions

- <For each FAIL, suggest the specific fix or relevant skill to run>
- <If type was defaulted, suggest: "Add Type metadata field to explicitly declare template type">

4. Pass/Fail Criteria (Type-Specific)

Full PRD & AI Product PRD:

  • •PASS — All required sections present, no placeholders, metadata complete, requirement IDs present. Non-critical checks (section order, enum completeness) may fail.
  • •FAIL — Any of these fail: missing required sections, missing metadata fields, placeholders present, no requirement IDs or priority labels

1-Pager:

  • •PASS — All 7 sections present (no tolerance since all required), no placeholders, metadata complete. Requirement ID/format checks are N/A.
  • •FAIL — Any section missing, missing metadata fields, or placeholders present

Shape Up Pitch:

  • •PASS — All 6 sections present (no tolerance since all required), Appetite field present, no placeholders, metadata complete. Requirement ID/format checks are N/A.
  • •FAIL — Any section missing, Appetite field missing, missing metadata fields, or placeholders present

General principle: Required section checks adapt based on detected type. Recommendation: Run /prd-analyze for deeper semantic quality analysis.

Operating Principles

  • •NEVER modify files — this is strictly read-only
  • •Keep it fast — use Grep for pattern matching, avoid reading the entire file line-by-line when possible
  • •This is a structural check, not a semantic one — do not evaluate content quality
  • •For deep quality analysis, recommend /prd-analyze instead
  • •Report line numbers for all issues so the user can navigate directly to problems