media-memory
Pass
Audited by Gen Agent Trust Hub on Nov 7, 2026
Risk Level: SAFECOMMAND_EXECUTIONINDIRECT_PROMPT_INJECTION
Full Analysis
- [COMMAND_EXECUTION]: The skill uses
uv runto execute local Python scripts (ingest.py,search.py) located in~/.claude/media-memory/scripts/. These commands are used for managing and searching the local vector database. - [INDIRECT_PROMPT_INJECTION]: The skill processes user-supplied media content and descriptions which are then stored and retrieved as part of the agent's context. It includes a proactive recall mechanism that surfaces relevant assets based on similarity.
- Ingestion points: Media files, screenshots, and natural language descriptions provided by the user or generated by the agent (found in
SKILL.md). - Boundary markers: The skill does not explicitly define markers to delimit retrieved media descriptions from active instructions.
- Capability inventory: The skill has the capability to write to a local asset directory and execute local search/ingest scripts via
uv run(found inSKILL.md). - Sanitization: There is no mention of sanitizing or escaping extracted text or descriptions before they are stored or surfaced to the agent.
Audit Metadata