The Agent Skills Directory

[COMMAND_EXECUTION]: The skill provides instructions and shell templates for the agent to execute curl commands. This is the intended mechanism for interacting with the Resemble AI REST API for media and text detection tasks.
[DATA_EXFILTRATION]: The skill transmits user-provided media files (audio, image, video) and text to app.resemble.ai. As this targets the vendor's official API for the primary purpose of the skill, it is categorized as legitimate data processing rather than malicious exfiltration.
[INDIRECT_PROMPT_INJECTION]: The skill acts as a surface for processing untrusted external content, including media files and text strings.
Ingestion points: Media provided via public HTTPS URLs, local file uploads (up to 150MB), and raw text inputs for analysis (SKILL.md).
Boundary markers: The skill implements the 'Iron Law'—a set of strict behavioral instructions that force the agent to rely solely on completed API results and numerical scores, preventing the agent from following instructions potentially embedded in the media style or tone.
Capability inventory: Capabilities include network requests via curl and file reads for multipart/form-data uploads (SKILL.md).
Sanitization: The instructions mandate the use of jq or Python-based JSON builders to sanitize user-provided text before it is interpolated into API request payloads, mitigating potential command or JSON injection risks.

resemble-detect