Onda adds agent-friendly manifest fields for Remotion components
Onda, an open-source library of motion graphic components for Remotion, has launched an update focusing on enhancing agent-native discovery for LLMs. This update introduces 'pickWhen' and 'composes' fields to manifest entries, which allows AI models to better select and combine components for video creation without extensive prompt engineering. The library now features 70 components and 18 transitions across various categories, all designed with a consistent motion language.
Key Takeaways
- `ComponentManifestEntry` now includes optional `pickWhen?: string` and `composes?: ReadonlyArray<string>` fields.
- The sync script passes those fields through from `<slug>.meta.json` and fails the build if a `composes` slug does not resolve.
- `summarizeRegistryAsMarkdown` now surfaces `*pick when:*` and `*composes:* [slug](#slug)` links when given the runtime manifest catalog.
- Onda’s catalog now spans 70 components across 8 categories plus 18 transitions.
- The update backfills `pickWhen` on 15 entrances and `composes` on 7 scenes per techspec appendices.
Why It Matters
The immediate effect is cleaner component selection for AI-driven video workflows: Onda is now encoding usage guidance in the manifest itself, rather than relying on prompt-side curation. That matters because the library is explicitly targeting Remotion users who want source-owned components and machine-readable discovery. The broader signal is that Onda is treating agent-native discovery as a first-class product surface, alongside its runtime manifest and markdown registry. The next concrete checkpoint is whether downstream consumers start reading the new `pickWhen` and `composes` entries in `llms.txt`, `llms-full.txt`, or the generated manifest output.
Read full article at github.com