The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill accepts user-influenced parameters such as --output-path and --window-name which are passed directly to the uloop CLI tool. This creates a surface for indirect prompt injection where malicious data could attempt to manipulate file output locations or target unintended windows. \n
Ingestion points: Parameters --output-path, --window-name, and --match-mode defined in SKILL.md.\n
Boundary markers: No explicit delimiters or instructions are provided to the agent to treat these parameters as untrusted data.\n
Capability inventory: The skill executes uloop record-video, which possesses the capability to write video files to the local file system.\n
Sanitization: The documentation mentions basic extension validation for .mp4 and .webm but does not specify path sanitization or validation of the window title strings.\n- [COMMAND_EXECUTION]: The workflow documentation instructs the agent to execute manual shell commands, specifically ffmpeg -i "<OutputPath>" .... If the tool returns a manipulated OutputPath (e.g., containing shell metacharacters like semicolons or backticks) and the agent does not perform strict shell escaping, this could lead to arbitrary command execution in the agent's host environment.

uloop-record-video