rejected-analysis
Pass
Audited by Gen Agent Trust Hub on Sep 15, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill ingests findings from synchronized logs which originate from external sources, potentially exposing the agent to adversarial instructions hidden within code-review data.
- Ingestion Points: Data is retrieved via 'scripts/larch.sh run-log sync' and 'prepare' commands as defined in SKILL.md.
- Boundary Markers: The workflow mentions a security-sensitive classifier and requires compliance with responsible disclosure guidelines.
- Capability Inventory: The skill is capable of launching agent-based reviews via 'agent launch-review' and filing public GitHub issues via the '/issue' skill.
- Sanitization: Verification occurs in read-only sandboxed environments, and findings are filtered by a Rust tool for security sensitivity before being processed for filing.
Audit Metadata