菜单
Validates documentation quality and freshness — checks for broken links, stale content, llms.txt sync, image issues, heading hierarchy, and badge URLs. Runs locally or in CI. Use to catch documentation decay before it reaches users.
用 Codex 或 Claude 帮你安装 复制这段 Prompt，粘贴到 Codex、Claude 或其他助手里，让它检查 Skill 页面并帮你完成安装。
基于 SOC 职业分类
Generate marketing-quality repository documentation from codebase analysis. Scans 10 signal categories, extracts features with file-level evidence, and produces README, CHANGELOG, ROADMAP, and 15+ more docs. Zero runtime dependencies. For AI context file management, see ContextDocs.
Systematic codebase scanning for features and evidence-based feature-to-benefit translation. Extracts what a project does from its code and translates it into what users gain — generates features and benefits sections, "Why [Project]?" content, and feature audit reports. Use when writing a features table for a README, extracting features from code, auditing feature coverage, or answering "why should someone use this project?".
Orchestrates documentation updates after version bumps, feature additions, or periodic maintenance. Analyses git history since the last release, identifies which docs are affected, and delegates to existing skills (changelog, feature-benefits, docs-verify, llms-txt, user-guides) for selective refresh. Delegates AI context updates to ContextDocs if installed. Use when releasing a new version or refreshing stale docs.
Generates READMEs with the Daytona/Banesullivan marketing framework — hero section, benefit-driven features, quickstart, comparison tables, and compelling CTAs. Produces docs that sell as well as they inform. Use when creating or overhauling a project README.
Generates ROADMAP.md from project milestones, issues, and boards (GitHub, GitLab, or Bitbucket). Structures content with mission statement, current milestone progress, upcoming milestones, and community involvement section. Use when creating or updating a project roadmap.
Generates task-oriented user guides and how-to documentation for a repository. Creates docs/guides/ with step-by-step instructions for common workflows, integrations, and advanced usage. Links guides into README.md and CONTRIBUTING.md. Use when a project needs user-facing how-to documentation beyond the README quickstart.
| name | docs-verify |
| description | Validates documentation quality and freshness — checks for broken links, stale content, llms.txt sync, image issues, heading hierarchy, and badge URLs. Runs locally or in CI. Use to catch documentation decay before it reaches users. |
| version | 1.5.0 |
Generating documentation is a solved problem. Preventing documentation decay is not. This skill validates that generated docs remain accurate, linked, and fresh over time.
Check heading hierarchy and structural consistency across all .md files. Verify: heading hierarchy (no H1 > H3 skips — critical for RAG/GEO), single H1 per doc, consistent formatting (trailing whitespace, list markers, blank lines around headings), no bare URLs.
Check all internal and external links. Internal: verify target file exists, anchor links match headings, check case-sensitivity (Linux vs macOS). External: check HTTP status, timeout 10s, skip authenticated URLs, flag 404s/5xx. Enhanced detection patterns (case-sensitivity, fragments, redirect chains, nested relative links) in SKILL-reference.md.
Verify llms.txt references match actual files on disk. Check for: missing files, orphaned docs not listed in llms.txt, description drift (first paragraph check), stale llms-full.txt.
Check all image references in Markdown. Verify: file exists on disk, alt text present (flag empty ![]()), absolute URLs for registry-published packages (see SKILL-reference.md for platform URL patterns), file size under 1MB.
Flag stale docs using git log last-modification dates. See SKILL-reference.md for per-file staleness thresholds. Compare against latest commit date, not calendar date — dormant projects with no commits shouldn't trigger warnings.
Compare README features against actual code using feature-benefits skill extraction. Flag undocumented features (code evidence exists, not in README) and over-documented features (claimed but no code evidence).
Verify shields.io badge URLs return valid responses (HTTP 200, not error SVGs). Flag broken badges and outdated URL formats.
After running all verification checks, calculate a numeric quality score. The score gives users a single number to track and improve — modelled on the grading approach used in documentation quality tooling across the ecosystem.
See SKILL-reference.md for the 6-dimension scoring rubric (Completeness, Structure, Freshness, Link Health, Evidence, GEO & Citation Readiness) and grade bands (A through F).
score = 100
for each check result:
apply deductions from SKILL-reference.md scoring dimensions
score = max(0, score)
grade = lookup(score) // A: 90–100, B: 80–89, C: 70–79, D: 60–69, F: <60
Report: show per-dimension breakdown and always include an actionable "To reach next grade" suggestion with the 1-2 highest-impact fixes. See SKILL-reference.md for the full report format example.
Supports ci argument for pipeline use (/docs-verify ci --min-score 70) and --min-score N threshold. See SKILL-reference.md for CI score export snippets (GitHub Actions, GitLab CI).
Verify docs/ files have valid YAML frontmatter per the user-guides skill standard. Required fields: title, description, type (one of tutorial, how-to, reference, explanation). Also check: title matches H1, related: paths exist on disk. Scoring: -2 per guide missing required frontmatter (Structure dimension).
Estimate token cost for all skill files in .claude/skills/ using wc -w × 1.3. Flag skills over 3,000 tokens (reference) or 5,000 tokens (combined). Full audit script and thresholds in SKILL-reference.md.
Scan docs for content that should never appear in public repos. Classify credential-pattern matches as: placeholder (acceptable), env var reference (acceptable), or real credential (block immediately). Also check for: leaked internal paths (/Users/, /home/, C:\Users\), internal hostnames/IPs (192.168., 10.0.), and non-existent package names (dependency confusion vectors).
Basic presence and staleness check for AI context files (CLAUDE.md, AGENTS.md, .cursorrules, etc.). Scoring: -2 per context file older than 90 days, -1 per missing file in the project's tool ecosystem. For full signal-gate scoring and line budget analysis, install ContextDocs.
When run with ci argument, output machine-readable ERROR:/WARN: lines with file:line format and exit code 1 on errors. Supports --min-score N threshold. Full CI output format and GitHub Actions workflow template in SKILL-reference.md.