The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted data from external URLs or local files and processes it through AI models to generate transcripts and scene descriptions. This content is then used to influence agent behavior or generate summaries.
Ingestion points: Untrusted media enters the system via coll.upload(url=...) and coll.upload(file_path=...) as documented in SKILL.md and reference/api-reference.md.
Boundary markers: There are no specific boundary markers or 'ignore' instructions provided when passing generated transcripts or scene descriptions to the agent or into coll.generate_text().
Capability inventory: The skill uses the videodb SDK to perform network operations (API calls to *.videodb.io) and the Bash tool to run Python scripts that write event logs to ~/.local/state/videodb/ via scripts/ws_listener.py.
Sanitization: The skill does not implement explicit sanitization of transcripts or AI-generated visual descriptions before processing them with further LLM calls.
[COMMAND_EXECUTION]: The skill requires the agent to execute shell commands to install dependencies and run local scripts for session management.
Evidence: SKILL.md provides instructions for pip install and running the scripts/ws_listener.py background process to capture real-time events.
[DYNAMIC_EXECUTION]: The skill heavily relies on the agent generating and executing Python snippets at runtime to interact with the VideoDB SDK.
Evidence: Multiple examples in SKILL.md and reference documents show the use of python -c and heredocs (python << 'EOF') to initialize connections and perform video operations.

videodb