implementing-zero-trust-with-hashicorp-boundary

Pass

Audited by Gen Agent Trust Hub on Mar 16, 2026

Risk Level: SAFECREDENTIALS_UNSAFEEXTERNAL_DOWNLOADSCOMMAND_EXECUTIONPROMPT_INJECTION
Full Analysis
  • [CREDENTIALS_UNSAFE]: The skill documentation includes configuration examples containing placeholder hardcoded credentials and encryption keys.
  • Evidence: SKILL.md contains a PostgreSQL connection string with a default password and AEAD kms blocks with sample keys.
  • [EXTERNAL_DOWNLOADS]: The deployment instructions involve downloading software components and GPG keys from trusted official repositories.
  • Evidence: SKILL.md directs users to fetch GPG keys and install the boundary package from HashiCorp's official apt repository.
  • [COMMAND_EXECUTION]: The provided Python auditing script programmatically invokes the boundary system CLI to gather infrastructure state.
  • Evidence: scripts/agent.py uses the subprocess module in run_boundary_cmd to execute discovery and audit commands.
  • [PROMPT_INJECTION]: The auditing tools ingest and process metadata from external CLI outputs which could contain malicious instructions if resource names or descriptions are compromised.
  • Ingestion points: JSON data returned from the boundary CLI in scripts/agent.py.
  • Boundary markers: No delimiters are used to wrap or sanitize ingested metadata before processing.
  • Capability inventory: Local system execution via subprocess and file system writes for reporting.
  • Sanitization: Output fields are parsed and printed directly without escaping or content validation.
Audit Metadata
Risk Level
SAFE
Analyzed
Mar 16, 2026, 03:43 PM