show-me-your-work
Pass
Audited by Gen Agent Trust Hub on Nov 6, 2026
Risk Level: SAFECOMMAND_EXECUTIONINDIRECT_PROMPT_INJECTION
Full Analysis
- [COMMAND_EXECUTION]: The skill includes a shell script
scripts/log.shused for appending data to a decision log. This script executes standard shell utilities likemkdirandprintf. - [INDIRECT_PROMPT_INJECTION]: The skill instructs agents to audit its logs against session transcripts and to spawn subagents for cross-model review of these materials, which creates a vulnerability surface for indirect instructions. 1. Ingestion points: The skill processes data from
agent-transcripts/and the generated TSV decision logs. 2. Boundary markers: The instructions do not define delimiters or markers to help models distinguish between logged data and potentially malicious embedded instructions. 3. Capability inventory: The skill relies on file writing via shell scripts and platform-level capabilities to spawn subagents for review. 4. Sanitization: Thescripts/log.shscript sanitizes spreadsheet formula characters (prefixing=,+,-, and@with a single quote) but does not provide sanitization for natural language instructions that could trigger behavior in the reviewing models.
Audit Metadata