The Agent Skills Directory

[PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection via the composition manifest file.
Ingestion points: The skill parses manifest.yaml to extract component paths and layout parameters in Step 1 and Step 2.
Boundary markers: No boundary markers or 'ignore' instructions are used when interpolating manifest data into shell commands.
Capability inventory: The skill executes Bash commands, yq, and ffmpeg subprocesses.
Sanitization: The values extracted from the manifest (such as components[].output) are used directly in shell loops and ffmpeg arguments without validation or escaping.
[COMMAND_EXECUTION]: The skill dynamically generates complex ffmpeg filter strings and command arguments based on the layout and options fields in the manifest. While these are intended operations, the lack of input validation on the manifest content could allow an attacker to craft a manifest that executes unintended ffmpeg features or influences the host file system.

media-composition