security-audit
Pass
Audited by Gen Agent Trust Hub on Oct 1, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTIONEXTERNAL_DOWNLOADS
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and analyze untrusted codebases as its primary function.
- Ingestion points: The skill processes external source code, API responses, and database schemas provided by the user for auditing purposes.
- Boundary markers: The skill implements a comprehensive 'Write isolation' and 'Artifact promotion' procedure in SKILL.md, ensuring that target-controlled code operates only within a network-less sandbox and cannot modify the agent context.
- Capability inventory: The skill utilizes local file system access and Git for audit management but explicitly prohibits target-controlled code from accessing external networks or host-level credentials.
- Sanitization: The skill provides specific 'Core discipline' instructions across its companion files (e.g., AI-AND-LLM.md, CLIENT-SIDE.md) to help the agent differentiate between untrusted data and system instructions.
- [EXTERNAL_DOWNLOADS]: The skill provides procedures for synchronizing its configuration and documentation from Cloudflare's official GitHub repository.
- [SAFE]: The skill employs defensive programming in its validator scripts (validate-findings.cjs, validate-coverage-ledger.cjs), using OS-level protections like no-follow descriptors to prevent directory traversal and race condition attacks.
Audit Metadata