show-me-your-work
Pass
Audited by Gen Agent Trust Hub on Nov 3, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTIONDYNAMIC_EXECUTION
Full Analysis
- [DYNAMIC_EXECUTION]: The skill invokes a local bash script
scripts/log.shto manage the decision log. This script implements defensive measures against spreadsheet formula injection (CSV injection) by prefixing inputs starting with characters like=,+,-, or@with a single quote. It also sanitizes inputs to ensure they remain single-line by stripping tabs and newlines. - [INDIRECT_PROMPT_INJECTION]: The auditing instructions require the agent to read and verify previous transcripts and session files (
history://,archive.sessions()). This constitutes an ingestion point for untrusted data that could contain adversarial instructions from previous interactions. However, the skill provides a structured framework for this audit, focusing on evidence verification (e.g., commit SHAs, PR numbers) which mitigates the risk of following injected instructions. - [DATA_EXPOSURE]: The skill accesses session logs for auditing purposes. The instructions include a specific security guideline advising the agent not to use broad search parameters (e.g.,
project: "*") when retrieving sessions, which helps prevent the accidental exposure of unrelated private interaction data.
Audit Metadata