The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill is designed to process external content from local directories and remote URLs, which could contain instructions to manipulate the agent's output or logic. \n
Ingestion points: The agent reads local source code files (README, styles, components) and fetches remote site content using a headless browser (SKILL.md). \n
Boundary markers: No specific boundary markers or 'ignore embedded instructions' delimiters are mandated for the processing of ingested data. \n
Capability inventory: The skill uses host-based command-line tools for media rendering and accesses the network to download website assets (SKILL.md). \n
Sanitization: The instructions include a comprehensive exclusion list for sensitive files (e.g., .env, credentials, .git) and prohibit including PII or secrets in the final video output. \n- [COMMAND_EXECUTION]: The skill requires the agent to invoke local command-line utilities (such as ffmpeg or similar tools) to perform video rendering and audio mixing tasks. \n- [DYNAMIC_EXECUTION]: The skill instructs the agent to dynamically build the video using 'whatever works on this machine', which implies generating and executing scripts for browser automation and media processing.

brag-slim