The Agent Skills Directory

[DYNAMIC_EXECUTION]: The script scripts/process_video.py uses the eval() function to calculate the frame rate (fps) from ffprobe metadata strings (e.g., '30/1'). This presents a potential vulnerability if a maliciously crafted video file causes the tool to return executable code within the metadata fields.
[COMMAND_EXECUTION]: The skill makes frequent use of subprocess.run() to invoke external utilities like ffmpeg, ffprobe, and lsof. While arguments are handled as lists, these operations provide the skill with extensive system interaction capabilities. Furthermore, scripts/browser_server.py binds a control server to 0.0.0.0, potentially exposing the browser automation interface to the local network.
[INDIRECT_PROMPT_INJECTION]: The skill records and extracts text from external websites, creating a surface for indirect prompt injection where malicious instructions on a page could influence the agent's actions.
Ingestion points: The _discover_elements method in scripts/browser_server.py extracts innerText and attribute values from the DOM.
Boundary markers: The snapshots sent to the agent do not include specific delimiters or instructions to distinguish untrusted web content from system prompts.
Capability inventory: The skill possesses capabilities for browser control (click, type, evaluate), file writing, and execution of media processing tools.
Sanitization: There is no evidence of sanitization for text content extracted from the browser before it is interpolated into the agent's context.

broll-browser-recorder