memory
Pass
Audited by Gen Agent Trust Hub on Sep 23, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTIONCOMMAND_EXECUTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill searches and processes external Markdown topic pages, creating a pathway for indirect prompt injection from untrusted data sources.
- Ingestion points: Operations in
recall.mdandmine-learn.mdingest content from external topic pages and session episodes. - Boundary markers: The skill mentions structural checks for metadata via the OKF profile, but does not specify delimiters to isolate content from the agent's instructions.
- Capability inventory: The skill uses
rgand has the ability to read and write files (drafts and topic pages). - Sanitization: Mandatory 'author-distinct' review is required before content is moved to permanent storage.
- [COMMAND_EXECUTION]: The skill uses a Python utility script,
scripts/toil-mining/recent_human.py, to process log data with parameters for time windows and session paths.
Audit Metadata