ketchup-video
Pass
Audited by Gen Agent Trust Hub on Oct 25, 2026
Risk Level: SAFECOMMAND_EXECUTIONINDIRECT_PROMPT_INJECTION
Full Analysis
- [COMMAND_EXECUTION]: The skill instructs the agent to "render one silent landscape video" (e.g., mp4) and explicitly states that "writing and rendering the video, outside the repo, is the one exception to its read-only rule." This requires the agent to execute system-level rendering tools or generate and run scripts (e.g., FFmpeg, Python) to create the media file.
- [INDIRECT_PROMPT_INJECTION]: The skill generates video content based on potentially untrusted external data such as diffs and recordings.
- Ingestion points: The 'Slides' section of
SKILL.mdspecifies using screenshots, recordings, diffs, and data produced by the work. - Boundary markers: None; the instructions do not include delimiters or warnings to ignore instructions that might be embedded within the files being summarized.
- Capability inventory: The skill can execute rendering commands, write files to the host system outside the repository, and load/read sibling skill files (e.g.,
../<name>/SKILL.md). - Sanitization: No sanitization, escaping, or filtering of the ingested data is described before it is used to create slide labels and content.
Audit Metadata