skills/docker/skills/docker-agent-run/Gen Agent Trust Hub

docker-agent-run

Pass

Audited by Gen Agent Trust Hub on Sep 22, 2026

Risk Level: SAFE
Full Analysis
  • Safety Mode Configuration: The skill defines several safety levels (strict, balanced, restricted, autonomous) for agent tool execution. It specifically recommends the restricted mode for unattended or CI environments to ensure unexpected tool calls are denied rather than executed without review.
  • Sandbox Isolation: The skill provides instructions for using the --sandbox flag to run agents in isolated microVMs. It correctly identifies the trust boundaries, noting that while shell and process activity are isolated, workspace changes and Git hooks may interact with the host environment.
  • Network Policy Management: It includes commands for managing the sandbox network allowlist (docker agent sandbox allow/deny), which follows a default-deny policy, enhancing the security posture of remote interactions.
  • Diagnostic Tooling: The use of docker agent doctor is documented as a way to verify credentials and model availability locally before execution, ensuring that sensitive configuration issues are diagnosed rather than resulting in silent failures.
  • Trusted External References: The skill references official Docker documentation and GitHub repositories under the docker organization for its rules and provenance, which are recognized as trusted sources.
Audit Metadata
Risk Level
SAFE
Analyzed
Sep 22, 2026, 08:46 AM
Security Audit — agent-trust-hub — docker-agent-run