The Agent Skills Directory

[COMMAND_EXECUTION]: The skill uses the Bash tool to execute ffmpeg and ffprobe for the purpose of analyzing video recordings and extracting frames. These commands are restricted to the primary goal of the skill (motion analysis) and use standard parameters.
[EXTERNAL_DOWNLOADS]: The skill employs WebFetch to retrieve HTML and CSS from user-specified URLs. This is a core functionality required to audit live websites and does not involve the execution of untrusted remote scripts.
[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted data from external sources, including website content and media files. This presents a theoretical attack surface for indirect prompt injection. However, the risk is assessed as minimal because the agent's tasks are highly structured (e.g., parsing CSS values or sampling hex codes) rather than open-ended instruction following from the input data.
Ingestion points: External URLs fetched via WebFetch and media files accessed via Read and Bash tools.
Boundary markers: Not explicitly defined in the logic blocks, though the pipeline structure inherently limits the interpretation of input as instructions.
Capability inventory: File system writes (tokens), network fetches (WebFetch), and local command execution (ffmpeg).
Sanitization: Standard tool-based parsing of CSS and media metadata acts as a natural filter for instruction-based payloads.

design-context-extract