visa-vulnerability-agentic-harness
Pass
Audited by Gen Agent Trust Hub on Jun 30, 2026
Risk Level: SAFEPROMPT_INJECTION
Full Analysis
- [PROMPT_INJECTION]: The skill documentation describes a tool that ingests untrusted code from local or remote repositories for analysis by LLMs. This functionality creates an inherent attack surface for indirect prompt injection, where malicious instructions embedded in the scanned code could attempt to influence the agent's behavior or the scan results.
- Ingestion points: Code repositories provided via the
--repoflag or bulk repository lists in CSV format via the--repo-fileargument (SKILL.md). - Boundary markers: Not explicitly defined in the provided documentation.
- Capability inventory: The tool executes file system writes to create reports and manifests, and performs network operations to communicate with AI provider APIs (Anthropic, OpenAI) (SKILL.md).
- Sanitization: No specific sanitization or escaping mechanisms for the ingested code are described in the setup guide.
- [SAFE]: The skill provides instructions for managing sensitive configuration data, such as API keys and mTLS certificates, using industry-standard practices like
.envfiles and environment variables. No patterns of data exfiltration or hardcoded credentials were detected.
Audit Metadata