show-me-your-work
Pass
Audited by Gen Agent Trust Hub on Sep 27, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTIONCOMMAND_EXECUTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill defines a workflow where a subagent reviews a decision log and transcript. This creates a multi-step chain where data processed in the first phase could potentially influence the subagent in the review phase.\n
- Ingestion points: The agent is instructed to read the conversation transcript to audit the log via the Skill tool.\n
- Boundary markers: The skill enforces a single-line TSV format and restricts the type of content allowed in the 'evidence' field, which acts as a structural boundary.\n
- Capability inventory: The skill utilizes a local shell script (scripts/log.sh) to write entries to the filesystem.\n
- Sanitization: The logging script (scripts/log.sh) proactively sanitizes inputs by removing newlines and tabs to maintain file structure and prefixes formula-triggering characters (=, +, -, @) with a single quote to prevent injection attacks in spreadsheet software.\n- [COMMAND_EXECUTION]: The skill relies on a local bash script (scripts/log.sh) to perform file operations. While the script is part of the skill and includes sanitization for the data it writes, it represents an execution point for local system commands.
Audit Metadata