The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted data from external websites and local project files, which could contain malicious instructions designed to influence the agent's behavior.
Ingestion points: The skill fetches live website content via a browser and reads local project source code files as documented in the "Inspect" section of SKILL.md.
Boundary markers: The instructions lack explicit delimiters or warnings to the agent to treat fetched content as data rather than instructions.
Capability inventory: The skill uses tools to write files to the local system, execute rendering commands, and perform network requests to fetch assets.
Sanitization: No specific sanitization or filtering logic is provided for the content gathered from external sources before it is processed.
[DYNAMIC_EXECUTION]: The skill requires the agent to generate and run scripts locally to handle the video creation process.
Evidence: The "Build, check, render" section instructs the agent to build the video using "whatever works on this machine," implying the creation and execution of scripts to interface with media tools.
[COMMAND_EXECUTION]: The skill relies on running local system commands to capture data and render video output.
Evidence: The use of a "headless browser" for capturing website state and the generation of an "mp4" file indicates the execution of command-line utilities.
[EXTERNAL_DOWNLOADS]: The skill performs network requests to download assets from arbitrary user-provided URLs.
Evidence: The "Website" section directs the agent to download logos, images, and other visual assets from the targeted site into a local working directory.

brag-slim