The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes data from external video URLs and the resulting API metadata (such as title and hook_sentence). This introduces a potential surface for indirect prompt injection if the source content contains instructions meant to influence the agent.
Ingestion points: The agent reads data from the muapi.ai API response, specifically the shorts array containing title and hook_sentence (found in SKILL.md under Output Schema).
Boundary markers: The instructions do not define clear delimiters or warnings for the agent to treat the API-returned content as untrusted data.
Capability inventory: The skill utilizes shell script execution (run-ai-clipping.sh) and interactions with the muapi-cli tool to perform network operations and file uploads.
Sanitization: There is no evidence of sanitization or filtering applied to the text fields returned by the API before they are presented to the user or processed further.
[COMMAND_EXECUTION]: The skill instructs the agent to execute a local shell script located at library/edit/ai-clipping/scripts/run-ai-clipping.sh. This script is responsible for resolving video paths, performing uploads, and invoking the muapi-cli to handle the clipping process.

muapi-ai-clipping