kinetograph-taste-clone
Pass
Audited by Gen Agent Trust Hub on Nov 8, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTIONEXTERNAL_DOWNLOADSCOMMAND_EXECUTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill implements a workflow where sub-agents (using the Sonnet model) are tasked with reading video frames and writing technical specifications. This creates an attack surface where malicious visual content (e.g., text embedded in a video) could attempt to influence the sub-agent's output or instructions. The skill mitigates this by having the main model merge the specs and use measurement scripts (
scripts/measure.py) to verify estimated values before they are used in the build process. - [EXTERNAL_DOWNLOADS]: The
scripts/fetch.shutility usesyt-dlpto download video content from arbitrary URLs provided by the user. The instructions also suggest using a browser-driver agent to fetch assets from public domain repositories. The skill includes built-in warnings to ensure that downloaded media is kept in git-ignored directories to prevent accidental leaks or copyright violations. - [COMMAND_EXECUTION]: Several Python and Shell scripts execute system-level commands to perform media processing.
scripts/measure.py,scripts/timeline.py, andscripts/trace.pyuse the Pythonsubprocessmodule to callffmpeg,ffprobe, andpotrace. These calls are implemented using list-style arguments rather than shell strings, which is a safer practice for handling file paths and user-provided inputs.
Audit Metadata