The Agent Skills Directory

[COMMAND_EXECUTION]: Capability discovery via subprocess execution. The scripts/discover-roster.py script probes for installed agent CLIs (e.g., claude, gemini, grok) using subprocess.run with list-based arguments to prevent shell injection.
[COMMAND_EXECUTION]: Environment-aware probing. The discovery script uses bash -ic to source the user environment for capability checks, ensuring agents have access to necessary authentication profiles. It uses shlex.join to safely construct these commands, maintaining security while verifying agent readiness.
[SAFE]: Human-gated orchestration model. The core workflow enforces strict boundaries for high-risk actions, requiring explicit human signoff for data writes, service restarts, and deployments. It mitigates autonomous risk by splitting tasks at security gates and mandating independent, cross-vendor reviews.

sop