videodb
Pass
Audited by Gen Agent Trust Hub on Jun 16, 2026
Risk Level: SAFEEXTERNAL_DOWNLOADSCOMMAND_EXECUTIONDATA_EXFILTRATIONPROMPT_INJECTION
Full Analysis
- [EXTERNAL_DOWNLOADS]: Installs the
videodbPython SDK andpython-dotenvpackage from the official Python Package Index (PyPI) to facilitate interaction with the VideoDB service. - [COMMAND_EXECUTION]: Executes the bundled
scripts/ws_listener.pyscript and utilizes inline Python commands via theBashtool to manage session metadata and process real-time events. - [DATA_EXFILTRATION]: Ingests and transmits media data from local files, public URLs, and live RTSP/RTMP streams to the VideoDB cloud infrastructure for processing and indexing as part of its core functionality.
- [PROMPT_INJECTION]: Potential indirect prompt injection surface detected:
- Ingestion points: Ingests external content from third-party URLs (
coll.upload(url=...)) and live streams (coll.connect_rtstream(...)) as documented inSKILL.mdandreference/rtstream-reference.md. - Boundary markers: Untrusted transcript data and visual descriptions are interpolated into prompts for the
coll.generate_textmethod without explicit delimiters in the provided usage examples. - Capability inventory: The agent has access to
Bashandpythontools, allowing it to execute the VideoDB SDK for file system operations and network requests. - Sanitization: No explicit sanitization or filtering of transcribed text or visual AI descriptions is performed before the content is passed back into the LLM context for summarization or analysis.
Audit Metadata