show-me-your-work
Pass
Audited by Gen Agent Trust Hub on Oct 2, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTIONCOMMAND_EXECUTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill instructs the agent to read session transcripts from the host's session store to audit and verify log entries. While this involves processing potentially untrusted data from user inputs or previous model outputs, the instructions include a specific privacy safeguard to prevent reading transcripts from unrelated workspaces.
- [COMMAND_EXECUTION]: The skill uses a local bash script
scripts/log.shto manage the TSV log file. This script demonstrates secure coding practices by sanitizing inputs to prevent CSV/Formula injection attacks. It ensures that cells containing characters like=,+,-, or@are prefixed with a single quote to prevent execution if the log is opened in spreadsheet software. - [DYNAMIC_EXECUTION]: The skill facilitates a self-review process by spawning a subagent on a different model family. This is an orchestrated multi-agent workflow intended for verification and does not involve the execution of arbitrary or untrusted code strings.
Audit Metadata