A Claude Code skill that evaluates developer documentation sites for autonomous agent consumption — not just whether an LLM can read the docs, but whether an AI coding agent can discover, retrieve, parse, and act on them to complete tasks independently.

The skill produces a professional Word document (.docx) report with scored dimensions, specific findings, improvement opportunities, and strategic recommendations. It supports both initial baseline assessments and follow-up comparative evaluations that measure improvement against a previous report.

agent-docs-audit/
├── SKILL.md                         # The skill definition (start here)
└── references/
    ├── report-structure.md          # Templates for baseline and comparative report formats
    └── scoring-rubric.md            # Detailed scoring criteria for all six evaluation dimensions

• SKILL.md -- The main skill file. Contains the evaluation framework, execution workflows for both baseline and comparative modes, the six scoring dimensions, and guidance for producing honest, evidence-based reports.
• references/report-structure.md -- Defines the structure and sections for both baseline reports and before/after comparative reports. Used during report generation to ensure consistent, professional output.
• references/scoring-rubric.md -- Detailed rubric for scoring each of the six dimensions on a 1-10 scale, with specific criteria and evidence expectations at each score level.

The skill scores documentation across six dimensions:

1. LLM Discoverability & Ingestion -- How easily can an agent find and load the right content?
2. Agent Task Completion -- Can an agent follow the docs to complete a real task?
3. Structured Data & Machine Parseability -- Is critical data extractable without NLP?
4. Error Handling & Edge Case Coverage -- Can an agent recover when things go wrong?
5. Token Efficiency -- Can an agent get what it needs without loading unnecessary content?
6. Context Linking & Dependency Graphs -- Can an agent understand relationships between docs?

Copy the agent-docs-audit/ directory into your Claude Code project skills folder, or reference it from your Claude Code configuration.

Ask Claude Code to evaluate a documentation site for agent readiness. The skill triggers on phrases like:

• "Audit [site] for agent-first readiness"
• "Evaluate how AI-ready [site]'s documentation is"
• "Assess [site]'s docs for LLM/agent consumption"
• "How well would an agent work with [site]'s docs?"

Claude will fetch the site, evaluate it across the six dimensions, and generate a .docx report with scores, findings, and improvement opportunities.

If the site has been improved since a previous assessment, provide the earlier .docx report and ask for a follow-up evaluation:

• "Re-evaluate [site] against the previous baseline report"
• "Measure improvement on [site]'s docs since the last assessment"

Claude will extract baseline scores from the previous report, re-evaluate the current state, and produce a before/after comparison report showing measurable improvement with evidence.

Both modes produce a professional .docx report suitable for sharing with executives, clients, or technical leadership. Reports include scored dimensions, specific evidence-based findings, prioritized improvement opportunities (with effort/impact estimates), and strategic recommendations.

An example output (converted to PDF) is included here showing an example output created by running the skill against Polkadot Documentation on April 4, 2026.