probe
Pass
Audited by Gen Agent Trust Hub on Oct 22, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTIONEXTERNAL_DOWNLOADS
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill is designed to process output from dynamic security scanners (ZAP, Nuclei, Burp) which naturally ingests data from untrusted external targets, creating an attack surface for indirect prompt injection.
- Ingestion points: The skill processes findings and logs from external security tools (referenced in
SKILL.mdandreference/sarif-integration.md). - Boundary markers: The instructions explicitly state 'Trust nothing. Report only what you can verify' and require manual validation of all findings.
- Capability inventory: The skill utilizes CLI tool execution (e.g.,
zap.sh,nuclei) and file system operations to generate reports (SARIF, HTML). - Sanitization: The methodology mandates human-in-the-loop validation and reproducible evidence before confirming any vulnerability.
- [EXTERNAL_DOWNLOADS]: The skill provides instructions for installing well-known security tools from established sources.
- Fetches the Nuclei scanner from ProjectDiscovery's official GitHub repository (
github.com/projectdiscovery/nuclei). - References the
zapv2Python library for interacting with the OWASP ZAP API. - [PROMPT_INJECTION]: Static analysis flagged instructions related to overriding system prompts; however, these are identified as documentation for security testing (e.g., testing LLM endpoints for prompt injection vulnerabilities as defined in OWASP LLM01) rather than a malicious attempt to subvert the agent's own instructions.
Audit Metadata