infra-security

Pass

Audited by Gen Agent Trust Hub on Jun 24, 2026

Risk Level: SAFEPROMPT_INJECTION
Full Analysis
  • [PROMPT_INJECTION]: The skill processes untrusted external Infrastructure-as-Code (IaC) manifests (Terraform, Kubernetes, CloudFormation, Docker), creating a surface for indirect prompt injection. Malicious instructions could be embedded within comments or metadata of scanned files to influence the auditor's judgment during Stage 2.
  • Ingestion points: The scripts/scan.py script reads files from a user-provided target directory, and the AI model is instructed to read these files directly for semantic analysis.
  • Boundary markers: No specific delimiters or "ignore embedded instructions" warnings are configured for the model when reading untrusted IaC content.
  • Capability inventory: The skill allows local file reading and model reasoning, but instructions explicitly forbid network access, credential reading, or code execution (e.g., terraform apply, kubectl apply).
  • Sanitization: While the Stage 1 scanner uses deterministic regular expressions (safe from injection), the Stage 2 semantic analysis consumes raw file content without sanitization.
Audit Metadata
Risk Level
SAFE
Analyzed
Jun 24, 2026, 11:36 AM
Security Audit — agent-trust-hub — infra-security