The Agent Skills Directory

[COMMAND_EXECUTION]: The skill is designed to interact with external agent runtimes and tools such as codex, ntm, and claude through a dispatch mechanism. It probes for the presence of these tools at runtime to determine available adapters.
[SAFE]: The architectural design focuses on security best practices for multi-agent systems, including role isolation (limiting Implementers to their declared subjects) and the use of fresh, read-only contexts for Validators to prevent corruption of evidence.
[SAFE]: The included Python script scripts/fake_model_runner.py is a utility for testing and conformance. It uses only standard library modules to generate mock JSON artifacts for integration tests and does not perform any network operations or unauthorized file access.

agent-native