The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes external materials (scripts, repositories, links) to generate video content, which presents an attack surface for instructions embedded in that data.
Ingestion points: Reads material from links, repositories, and documents as described in Step 1 of SKILL.md.
Boundary markers: The instructions do not define clear delimiters or specific guidance to ignore instructions embedded in the ingested content.
Capability inventory: The skill can execute shell commands (ffmpeg, npx) and write files to the local environment during the scaffold and build phases.
Sanitization: No explicit sanitization or validation of the text/quotes extracted from the material is performed before they are used in the generated assets.
[COMMAND_EXECUTION]: The skill utilizes system commands and subprocesses for audio mixing, image snapshots, and video rendering.
Evidence: SKILL.md contains instructions to run npx, chmod, and node. build-audio.mjs invokes ffmpeg using execFileSync. finish.sh also uses ffmpeg and ffprobe for media processing.
[DYNAMIC_EXECUTION]: The audio build script executes the contents of a configuration file at runtime to load its data structures.
Evidence: template/build-audio.mjs uses Node's vm.runInNewContext to execute assets/cues.js. This is used to load the CUES object into the script's execution context for processing audio timing.

motion-film