system-review

name	system-review
description	Systematic 7-dimension review of a batch of PRD documents with multiple output modes. Use when (1) conducting a full system-level review across business value, architecture, competition, product strategy, tech evolution, PM assessment, and action planning, (2) generating next-direction recommendations for a requirement set, (3) evaluating PM writing quality and product thinking from requirement documents, (4) drafting a new PRD based on historical analysis of related documents, (5) producing any of these outputs: full_report, next_directions, quality_assessment, prd_draft. Also use when a user mentions 'system review', '7-dimension review', 'PM assessment', 'requirement review report', 'quality evaluation', or 'PRD draft generation'.
license	MIT
compatibility	opencode

System Review

Systematic 7-dimension review of a batch of PRD documents, producing different output types from the same foundational analysis.

The core insight: all output types share the same 7-dimension analysis — only the final formatting and prompt layer differ.

7-Dimension Review Framework

The dimensions execute in order because later dimensions depend on earlier ones:

Order	Dimension	Core Question	Prompt File
1	Business Value	What business problem does this set solve? Strategic value?	business-value.md
2	Architecture	Are the document classification, evolution stages, and dependencies reasonable?	architecture.md
3	Competition	Differentiation advantages and gaps vs competitors?	competition.md
4	Product Strategy	Is the product roadmap and prioritization reasonable?	product-strategy.md
5	Tech Evolution	Is the technical approach evolving reasonably? Any tech debt?	tech-evolution.md
6	PM Assessment	PM writing style and product thinking strengths and blind spots?	pm-assessment.md
7	Action Plan	What should be done in short/mid/long term?	action-plan.md

Dependency chain: 1→2→3→4→5→6→7. Each dimension receives the output of all prior dimensions as context.

Output Modes

                        ┌─ --output-type full_report ──→ Full 7-dimension review report (MD)
                        │
7-dimension analysis ───┼─ --output-type next_directions ──→ Next requirement direction recommendations (MD)
                        │
                        ├─ --output-type quality_assessment ──→ PM writing quality assessment (MD)
                        │
                        ├─ --output-type prd_draft ──→ New PRD draft based on historical analysis (MD)
                        │
                        └─ --output-type all ──→ All outputs

Output Type	Reference Case	Description
full_report	智能联动 960-line review	Complete 7-dimension report with action plan and milestones
next_directions	服务预约 next-step suggestions	What requirements to write next
quality_assessment	服务预约 PRD quality evaluation	PM writing and thinking assessment with scoring
prd_draft	声纹管理 V3 PRD (621 lines)	New PRD draft generated from historical analysis

--target-doc: When specified, generates an additional context report for a specific target document (historical evolution, related boundary issues, existing solutions). Useful when preparing to review a new PRD.

Dual Engine Architecture

Engine	Purpose	Default Model	When Used
text_engine	All 7 dimensions analysis	claude-sonnet-4-20250514	Always
vision_engine	Image understanding in original docs	claude-sonnet-4-20250514	When --enable-vision

Override via environment variables:

• TEXT_MODEL / VISION_MODEL — override default models
• ANTHROPIC_API_KEY — required

Quick Start

All commands below assume the working directory is the skill root (skills/system-review/). Install dependencies first: pip install -r requirements.txt

Full system review

python scripts/review.py <classify_json> <analysis_dir> <output_json> [options]

# Full 7-dimension review report
python scripts/review.py classify.json ./analysis/ result.json --output-type full_report

# Next directions only
python scripts/review.py classify.json ./analysis/ result.json --output-type next_directions

# PM quality assessment
python scripts/review.py classify.json ./analysis/ result.json --output-type quality_assessment

# Generate PRD draft for a specific document
python scripts/review.py classify.json ./analysis/ result.json --output-type prd_draft --target-doc doc123

# All output types
python scripts/review.py classify.json ./analysis/ result.json --output-type all

# With vision for original document images
python scripts/review.py classify.json ./analysis/ result.json --output-type full_report --enable-vision

# Specific dimensions only
python scripts/review.py classify.json ./analysis/ result.json --dimensions 1,6,7

# With industry context for competition dimension
python scripts/review.py classify.json ./analysis/ result.json --industry smart_home

PM assessment only (standalone)

python scripts/pm_assess.py <classify_json> <analysis_dir> <output_json> [options]

# Basic PM assessment
python scripts/pm_assess.py classify.json ./analysis/ result.json

# With custom scoring rubric
python scripts/pm_assess.py classify.json ./analysis/ result.json --rubric templates/pm-scoring-rubric.json

Input

Required

1. Classify result JSON — output from prd-overview-classify (categories, version chains, dependencies, document paths)
2. Analysis directory — directory containing per-document analysis JSONs from prd-per-analysis

Optional

• Industry template — via --industry flag (e.g., smart_home) for competition dimension context
• Competition references — via --competition-refs for user-provided competitor information
• Scoring rubric — via --rubric for PM assessment dimension override
• Review context — via --review-context for project-level specifications (scoring rubrics, domain rules, writing standards)

Review Context Injection

The --review-context parameter accepts a JSON file with project-level specifications that override defaults:

{
  "context_version": 3,
  "specifications": [
    {"type": "scoring_rubric", "content": "..."},
    {"type": "domain_rules", "content": "..."},
    {"type": "writing_standard", "content": "..."}
  ],
  "scoring_overrides": {
    "writing_dimensions": [
      {"name": "逻辑结构", "weight": 0.3},
      {"name": "技术深度", "weight": 0.25}
    ],
    "thinking_dimensions": [
      {"name": "迭代思维", "weight": 0.3},
      {"name": "体验思维", "weight": 0.25}
    ]
  }
}

Output Format

The output JSON structure varies by --output-type:

{
  "project_name": "智能联动",
  "output_type": "full_report",
  "dimensions": {
    "business_value": { "...": "..." },
    "architecture": { "...": "..." },
    "competition": { "...": "..." },
    "product_strategy": { "...": "..." },
    "tech_evolution": { "...": "..." },
    "pm_assessment": { "...": "..." },
    "action_plan": { "...": "..." }
  },
  "reports": {
    "full_report_md": "... (Markdown content)",
    "next_directions_md": "... (optional, when output-type includes it)",
    "quality_assessment_md": "... (optional)",
    "prd_draft_md": "... (optional, when --target-doc specified)"
  },
  "metadata": {
    "total_docs": 29,
    "dimensions_executed": [1, 2, 3, 4, 5, 6, 7],
    "models_used": {"text": "claude-sonnet-4-20250514", "vision": "..."}
  }
}

PM Assessment Details

The PM assessment dimension is the most unique and valuable sub-module.

Writing Style (4 dimensions × 1-5 score)

Dimension	5-point	3-point	1-point
Logic Structure	Clear flowcharts, state machines, decision rules per requirement	Basic structure but not systematic	No structure, scattered descriptions
Tech Depth	Accurate SDK versions, APIs, tracking plans, algorithm params	Basically correct but lacks detail	Vague or incorrect
Boundary Awareness	Explicit scope and exclusions per document	Some documents have boundaries	No boundary definitions
Business Perspective	ROI analysis, quantified user value, business metrics	User value described but not quantified	Technical descriptions only

Product Thinking (4 dimensions × 1-5 score)

Dimension	5-point	3-point	1-point
Iteration Thinking	Clear problem evolution chains between requirements	Iteration exists but evolution logic unclear	Isolated requirements, no iteration
Experience Thinking	Interaction designed from user cognition perspective	Basic UX consideration	Pure feature description
Data Thinking	Data statistics and evaluation plans with key metrics	Data mentioned but no concrete plan	No data-related content
Business Thinking	Closed-loop business design with ROI and success criteria	Business thinking but not closed-loop	No business consideration

PM Type Classification

• All writing scores ≥ 4 AND business ≤ 2 → Technical PM
• Business thinking ≥ 4 AND tech depth ≤ 2 → Business PM | Relatively balanced → Balanced PM

Growth Path Generation

Based on blind-spot dimensions, generate short/mid/long term action items:

• Short-term (1-3 months): Address weakest dimension basics
• Mid-term (3-6 months): Build cross-dimension connection ability
• Long-term (6-12 months): Advance to higher role

Key Features

• Same analysis, different outputs: One 7-dimension pass produces all output types
• Ordered dimension execution: Later dimensions receive all prior dimension outputs as context
• Standalone PM assessment: pm_assess.py can be called independently
• Configurable industry context: Competition dimension uses industry templates
• Review context injection: Project-level specifications override defaults
• Targeted document support: --target-doc generates context for a specific new PRD

Dependencies

pip install -r requirements.txt
# Core: anthropic, pydantic

Scripts Reference

Script	Purpose
review.py	Main entry: 7-dimension review with output type selection
pm_assess.py	Standalone PM assessment (dimension 6 only)

Prompts Reference

Prompt	Purpose
system-context.md	System context injected into all LLM calls
business-value.md	Dimension 1: Business value analysis
architecture.md	Dimension 2: Requirement architecture review
competition.md	Dimension 3: Competitive positioning
product-strategy.md	Dimension 4: Product strategy evaluation
tech-evolution.md	Dimension 5: Technical evolution review
pm-assessment.md	Dimension 6: PM capability assessment
action-plan.md	Dimension 7: Action plan and prioritization

For detailed API and customization, see references/usage-guide.md.