The Agent Skills Directory

[PROMPT_INJECTION]: The skill exhibits a surface for indirect prompt injection by processing untrusted data from video transcripts and visual descriptions.\n
Ingestion points: Untrusted data enters the agent's context through methods like video.get_transcript_text() and event logs stored in /tmp/videodb_events.jsonl.\n
Boundary markers: There are no explicit markers or instructions to the agent to ignore potentially malicious commands embedded within the transcript text.\n
Capability inventory: The agent can execute Python code using the videodb SDK via the Bash tool, allowing for media manipulation and API interactions.\n
Sanitization: The skill does not describe any sanitization or validation of transcript content before it is interpolated into prompts for analysis.\n- [EXTERNAL_DOWNLOADS]: The skill recommends installing the videodb SDK and python-dotenv. These are official vendor packages and well-known utilities.\n- [COMMAND_EXECUTION]: Functionality relies on the execution of Python code through the Bash tool to interact with the VideoDB API, which is the primary intended behavior of the skill.\n- [DATA_EXFILTRATION]: The provided ws_listener.py script saves WebSocket event data to local files in the /tmp directory (/tmp/videodb_events.jsonl). This is used for local session monitoring and does not involve sending sensitive data to unauthorized external endpoints.

videodb