verifiable-surfaces

Pass

Audited by Gen Agent Trust Hub on Sep 4, 2026

Risk Level: SAFEINDIRECT_PROMPT_INJECTIONDYNAMIC_EXECUTIONEXTERNAL_DOWNLOADS
Full Analysis
  • [INDIRECT_PROMPT_INJECTION]: The skill's 'Verification Ladder' requires the agent to process various application configuration files (e.g., Ansible, Terraform, Kubernetes manifests) and run them through native loaders or runtime smokes. This creates a surface where malicious instructions in config files could exploit the tools used for verification.\n
  • Ingestion points: Processes YAML, JSON, TOML, and infrastructure-as-code manifests as described in SKILL.md.\n
  • Boundary markers: Relies on third-party tool parsers; no specific delimiters are defined in the skill instructions to prevent instruction leakage from data.\n
  • Capability inventory: Involves execution of ansible-playbook, terraform, kubectl, docker compose, uv, and npm subprocesses.\n
  • Sanitization: Relies on the external tools' native loaders and does not perform independent sanitization of input files.\n- [DYNAMIC_EXECUTION]: The skill utilizes shell heredocs in references/config-examples.md to generate and immediately execute small Python and Node.js scripts. These scripts are used for verifying that configuration modules can be loaded successfully in an isolated environment.\n- [EXTERNAL_DOWNLOADS]: The skill fetches information using tools like npx (e.g., for shadcn) and provides links to official documentation for Ansible and Chezmoi.
Audit Metadata
Risk Level
SAFE
Analyzed
Sep 4, 2026, 06:53 AM
Security Audit — agent-trust-hub — verifiable-surfaces