The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted input such as code diffs, screenshots, and recordings to generate summaries and video slides. Malicious content within these inputs could influence the agent's output or rendering process. * Ingestion points: screenshots, recordings, diffs, and data (SKILL.md) * Boundary markers: None identified * Capability inventory: Video rendering and file writing outside the repository (SKILL.md) * Sanitization: None identified.
[COMMAND_EXECUTION]: The skill requires the agent to render a video file, which typically involves executing external CLI tools like ffmpeg or similar media processing utilities. * Evidence: "Render one silent landscape video, such as an mp4, with motion between slides." (SKILL.md).
[PROMPT_INJECTION]: The skill contains instructions that explicitly command the agent to bypass its operational constraints to enable file writing. * Evidence: "Writing and rendering the video, outside the repo, is the one exception to its read-only rule." (SKILL.md)

ketchup-video