Repository navigation
TanStackIntent
AI coding agents often guess at how to use a library from stale training data or generic documentation, and the result is code written against the wrong version or the wrong idiom. TanStack Intent addresses this by letting a library ship its own agent guidance, in the form of Agent Skills, inside the same npm package as the code. Because the guidance travels with the package, the agent reads instructions that match the version actually installed.
The tool is a command-line utility for library maintainers to generate, validate, and ship Agent Skills alongside their npm packages, and it follows the open Agent Skills standard. On the consumer side, it scans workspace dependencies for static skill files without executing package code. An allowlist and exclusions control which packages may contribute guidance, and the agent loads a skill only when a task needs it. Freshness is handled by declaring the source documents each skill depends on: the intent stale command flags a skill when those sources change, and a maintainer decides whether the skill needs an edit.
The project is marked alpha, with documentation at version 0. The repository was created in January 2026, is MIT licensed, and at the time of writing has 350 stars and 14 open issues.
TanStack Intent sits in the Tools quadrant because it is a discrete CLI used by maintainers and consumers to package and discover agent guidance, rather than a hosted service or a general technique.
The ring is Assess. The tool is alpha, and its star count and issue activity describe external interest, not confirmed use. No one on this radar has yet shipped a skill in a package or consumed one from a dependency. A move to Trial would require doing either in a real project at least once.
