show-me-your-work
Pass
Audited by Gen Agent Trust Hub on Sep 12, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill facilitates the ingestion of external task-related data (such as PR titles and artifacts) to generate a decision log. This processing of potentially untrusted data presents a surface for indirect prompt injection.
- Ingestion points: Data enters the agent context through task history APIs and conversation digests as described in the
SKILL.mdinstructions. - Boundary markers: The instructions do not define specific delimiters for separating untrusted data from the agent's internal logic, although the TSV output structure provides a degree of formatting.
- Capability inventory: The skill utilizes
scripts/log.shto perform filesystem operations including file writing (printf >>) and directory creation (mkdir -p). - Sanitization: The accompanying script
scripts/log.shimplements security controls to strip control characters and prefix characters that trigger formula execution in spreadsheet applications, reducing the risk of downstream exploitation during manual review.
Audit Metadata