source-driven-development

Use source-driven development: detect exact versions from package.json or pyproject.toml, fetch the relevant official docs page, implement the documented pattern, and cite the source URL. This prevents deprecated APIs from stale training data entering your codebase.

Fetch the specific documentation page for the feature, such as react.dev/reference/react/useActionState, rather than the homepage. Extract API signatures, deprecation warnings, and migration notes, then implement exactly what the docs show for your detected version.

Official documentation ranks first, followed by official blogs and changelogs, then web standards references like MDN, then compatibility data like caniuse.com. Stack Overflow answers, blog tutorials, and AI training data are never treated as primary sources.

The conflict is surfaced to you with both options explained, such as using the modern documented pattern versus matching the existing codebase convention. The Skill never silently picks one approach without your decision.

Skip it when correctness does not depend on a framework version, such as renaming variables, fixing typos, or pure logic like loops and data structures. It is also skipped when you explicitly prioritize speed over verification.

No. Fetched pages are treated as untrusted data: only API definitions, examples, and migration notes are extracted. Directives embedded in fetched content targeting the model are ignored, and outbound endpoints from docs are never hardcoded without disclosure.