In-depth guides

DocsCI extracts fenced code blocks from Markdown files and runs them in language-specific sandboxes. Python uses Pyodide WASM; JavaScript/TypeScript uses isolated-vm. A snippet passes if it exits cleanly (no exception, no uncaught rejection). Failures produce findings with the error message, line number, and an AI-generated fix.

• •Supported: python, javascript, typescript
• •Execution timeout: 20s default, configurable up to 60s
• •Findings include: error message, stdout, stderr (redacted), language, line number
• •AI fix: generated for all error-severity snippet failures

DocsCI parses your OpenAPI spec (YAML or JSON) and cross-references documented endpoints and parameters against mentions in your Markdown docs. Mismatches produce drift findings.

• •Detects: renamed parameters, removed endpoints, changed types
• •Supports: OpenAPI 3.x YAML and JSON
• •Configure: provide openapi_path (in repo) or openapi_url (remote)
• •Coverage bar: shows % of spec endpoints mentioned in docs

Docs are parsed as HTML via jsdom and checked with axe-core for WCAG 2.1 violations. Structural checks verify heading hierarchy, alt text, link text, and color contrast.

• •Rules: axe-core WCAG 2.1 Level AA + structural checks
• •Findings: missing alt text, empty links, heading skip, contrast
• •Severity: error (violations), warning (best practices)

DocsCI lints prose for quality signals that correlate with poor developer experience: passive voice, weasel words, hedging, sensitive terms, and Flesch-Kincaid grade level above 12.

• •Passive voice detection (heuristic + common irregular past participles)
• •FK grade level: warns if > 12 (target: 8–10 for technical docs)
• •Weasel words: 'very', 'quite', 'somewhat', 'basically', etc.
• •Hedging: 'might', 'could', 'perhaps', 'maybe'
• •Sensitive terms: customizable blocklist