The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and process potentially untrusted content from external sources, which could contain malicious instructions. • Ingestion points: Untrusted data enters the agent context through clipforge_ingest_product (URL), clipforge_product_script (URL/image), and clipforge_transcript_inspect (media transcripts). • Boundary markers: The skill includes a 'Hard Rule' (Rule 7) instructing the agent to treat fetched content strictly as data and never as instructions, providing a behavioral boundary. • Capability inventory: The skill has extensive capabilities including triggering video composition, searching stock footage, and managing local media assets via an API. • Sanitization: No technical sanitization or filtering of the external text is mentioned before it is processed by the language model.
[COMMAND_EXECUTION]: The skill documentation provides instructions for executing commands on the host machine to operate the video production toolchain. • Evidence: Documentation references the use of node bin/clipforge.mjs for CLI operations and pnpm dev or pnpm start to run the local API server. • Evidence: The skill requires and interacts with the ffmpeg binary for media composition.
[EXTERNAL_DOWNLOADS]: The skill involves the installation of external components and the fetching of remote media assets. • Evidence: The skill suggests installation using npx skills add xixihhhh/clipforge. • Evidence: The video pipeline fetches stock footage and metadata from external providers through the clipforge_search_stock tool.

clipforge-video