backend-security

Pass

Audited by Gen Agent Trust Hub on Nov 9, 2026

Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
  • [INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted external data, specifically the target project's source code, git logs, and configuration files, to conduct its security review. This creates a surface for indirect prompt injection where malicious instructions hidden in the analyzed files could attempt to influence the agent's behavior.
  • Ingestion points: SKILL.md (Steps 1, 3, 4, and 6) specifies reading repository source code and git history.
  • Boundary markers: The instructions do not define specific delimiters or instructions for the agent to ignore potentially adversarial content embedded within the files being triaged.
  • Capability inventory: The skill possesses the capability to execute local scripts (security-scan.js, checkpoint.js) and host-level CLI tools (gitleaks, npm audit, semgrep, etc.) via subprocess calls.
  • Sanitization: While the skill mandates the redaction of secret values in its final reports, it does not include instructions for sanitizing or escaping the raw source code content that it interpolates into its internal context during analysis.
Audit Metadata
Risk Level
SAFE
Analyzed
Nov 9, 2026, 09:36 AM