Doc2AHP

Turning academic decision theory into a practical tool for developers

中文

Doc2AHP is a Claude Code Skill that brings structured multi-criteria decision analysis to software engineering, based on the principles from the paper "Doc2AHP: Inferring Structured Multi-Criteria Decision Models via Semantic Trees with LLMs".

Tech stack selection, architecture decisions, library comparisons — these decisions often:

• Involve multiple conflicting evaluation dimensions
• Rely on gut feeling rather than systematic analysis
• Lack traceable decision records

Doc2AHP uses the AHP (Analytic Hierarchy Process) framework to transform vague "feelings" into quantifiable, traceable, structured decisions.

Doc2AHP supports two ways to begin a decision analysis:

• Mode A: Document-Grounded — Extract criteria from actual documents (requirements docs, tech specs, web search results) with full source traceability. Faithful to the paper's "Doc → AHP" pipeline. See example
• Mode B: Quick Analysis — Generate criteria from LLM domain knowledge for rapid exploration. Faster, but less traceable.

Copy the skill/doc2ahp-decision/ directory into your project's .claude/skills/:

# In your project root
mkdir -p .claude/skills
cp -r path/to/doc2ahp/skill/doc2ahp-decision .claude/skills/

Describe your decision scenario in Claude Code. The skill will guide you through the decision process:

0. Input Mode Selection — Choose Document-Grounded or Quick Analysis mode
1. Framework Construction — Define goal, extract criteria, build hierarchy
2. Multi-Perspective Evaluation — Evaluate from technical, business, ops perspectives
3. Consensus Aggregation — Merge perspectives into consensus weights
4. Consistency Check — Ensure logical consistency of judgments
5. Alternative Scoring — Weighted scoring to produce rankings
6. Decision Report — Structured, archivable report output

• Zero Dependencies: Pure Skill approach — the LLM itself is the AHP computation engine
• Cognitive Constraints: Max 7 criteria per level, max 3 depth levels (Miller's Law)
• Multi-Perspective Evaluation: Simulates the paper's multi-agent mechanism to reduce single-viewpoint bias
• Consistency Verification: Transitivity checks ensure logical coherence

doc2ahp/
├── skill/doc2ahp-decision/   # Claude Code Skill (core)
├── docs/methodology.md        # Detailed methodology
├── paper/PAPER_SUMMARY.md     # Paper summary
└── examples/                  # Usage examples

Good fit: 3+ alternatives, multi-dimensional trade-offs, architecture-level decisions, need to explain rationale to team

Not ideal: Simple A/B choices, single-dimension decisions, urgent decisions

• Methodology — Full mapping from paper to practice
• Paper Summary — Core concepts from the Doc2AHP paper

If the skill / methodology / examples in this project help your work, please cite the original paper:

@article{Wu2026Doc2AHP,
    author    = {Hongjia Wu and Shuai Zhou and Hongxin Zhang and Wei Chen},
    title     = {Doc2AHP: Inferring Structured Multi-Criteria Decision Models via Semantic Trees with LLMs},
    journal   = {arXiv preprint arXiv:2601.16479},
    year      = {2026},
    doi       = {10.48550/arXiv.2601.16479}
}

MIT