AI-Readable Media Sidecars

CDAF introduces an open sidecar format that stores descriptive information about video so agents can reuse prior analysis instead of repeatedly processing the same footage. The repository includes a specification, command-line tool, agent skill and reproducible benchmark, making this more concrete than a metadata proposal. If adopted, media pipelines could gain a machine-readable cache layer for cost, provenance and interoperability. Confirmation requires support from another media tool, measurable savings on real archives or adoption of the format outside its originating project.