DocSutra AIDocumentation Truth Audit - v0.2.1 - by Vatsal Shah

Roadmap and FAQ

What is planned (clearly separated from what exists) and answers to common questions.

Roadmap (not shipped)

NextItemWhy
1Opt-in external link checking with rate limits and SSRF protectionCloses the biggest "not checked" gap
2Full CommonMark/MDX AST parser, same test corpusFewer false negatives
3Git-diff driven impact mapping (which docs does a source change affect)Real drift detection
4Real CLI --help snapshot ingestion (user-approved execution)Replaces declared contracts
5OpenAPI/TypeDoc comparison adaptersAPI documentation truth
6Sandboxed snippet runner with policy filesVerify examples by running them safely
7Host activation smoke tests for the five IDEsTurn "installed" into "verified"
8MCP server and optional dashboardEasier assistant integration

Nothing above is available today. See Status.

FAQ

Does it need an AI key or internet? No. The scan is fully offline.

Will it run my README commands? Never. Commands are classified and compared to a declared contract only.

Does a green report mean my docs are correct? No. It means the listed checks passed. Read the ledger and the not-checked section.

Why not use a general link checker? Many ignore heading anchors, case sensitivity, accessibility and credentials, and none produce a ledger of what was and was not verified.

How accurate is it? 100% precision and recall on the author's labelled corpus; unknown on arbitrary repositories. Report misses as issues.

Does it support MDX, AsciiDoc or reStructuredText? MDX text is scanned as Markdown with limits; the others are not supported. See the coverage matrix.

Is it secure to run on untrusted repositories? The scan is read-only and never executes content, which is the design goal and is tested. It has not had a third-party audit.

Can I use it commercially? Yes, under Apache-2.0.

Where do I report problems? GitHub Issues for bugs; private vulnerability reporting for security.

Glossary

TermMeaning
AnchorThe #fragment of a link, matched to a heading slug
LedgerThe table of checks performed, passed and failed
BaselineAn earlier audit.json used to find only new problems
CLI contractSubcommands and flags you declare for a tool
SARIFStandard format for static-analysis results used by GitHub code scanning
Finding IDStable hash identifying one finding across runs