The Agent Skills Directory

[COMMAND_EXECUTION]: The skill relies on the execution of local command-line tools to perform its primary functions.
Executes cmux shot and cmux record to capture visual data from the application windows.
Mentions the execution of a Python script scripts/pr-media.py to handle media uploads and format conversion.
[INDIRECT_PROMPT_INJECTION]: The skill provides an interface for injecting arbitrary text into captured media, creating a potential surface for indirect prompt injection if the output is later processed by an agent with vision capabilities.
Ingestion points: Text inputs supplied via --label, --caption, and the cmux record note command are ingested and rendered into the resulting media files.
Boundary markers: None. Text is drawn directly into the screenshots or video frames.
Capability inventory: The skill has the capability to execute CLI commands (cmux), write files to user-specified paths (--out), and execute scripts that perform network uploads (scripts/pr-media.py).
Sanitization: The provided documentation does not mention sanitization or validation of the text inputs before they are rendered into the captures.
[DATA_EXFILTRATION]: The skill facilitates the movement of data from the local environment to external locations.
Documentation in SKILL.md indicates that scripts/pr-media.py is used to upload the captured screenshots and recordings, which constitutes a documented data exfiltration path for the purpose of sharing evidence in pull requests.

cmux-capture