retrospect-domain
Pass
Audited by Gen Agent Trust Hub on Oct 30, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTIONCOMMAND_EXECUTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill processes historical session data which may contain untrusted content from past interactions that could theoretically influence the agent's current analysis.\n
- Ingestion points: Conversation turns and tool usage history are read from session files in Step 3 of the analysis process.\n
- Boundary markers: No explicit delimiters or instructions are provided to isolate the untrusted session content from the analytical instructions.\n
- Capability inventory: The skill uses
Bash,Read,Write, andGreptools to execute local scripts and perform file operations.\n - Sanitization: No evidence of content sanitization or validation of the session data was found prior to model analysis.\n- [COMMAND_EXECUTION]: The skill executes an internal bash script,
retrospect-load-sessions.sh, using the Bash tool and passes user-provided arguments directly to it. This is a core part of the session filtering workflow.
Audit Metadata