The Agent Skills Directory

[COMMAND_EXECUTION]: The skill generates shell command templates for common infrastructure tools including docker, kubectl, and curl. These are intended to be executed by an operator to manage containerized inference workloads.
[DYNAMIC_EXECUTION]: The workflow dynamically constructs deployment runbooks by interpolating user-provided inputs, such as Hugging Face model IDs and port configurations, into shell script templates for Docker and Kubernetes.
[INDIRECT_PROMPT_INJECTION]: The skill possesses an attack surface for indirect prompt injection as it ingests untrusted external data (model IDs, hardware specifications, and benchmark parameters) to generate its outputs.
Ingestion points: Model identifiers, hardware inventory details, and benchmark load shapes provided by the user (SKILL.md).
Boundary markers: The skill explicitly uses an 'observed-command-gated' quality tier, requiring that commands be verified rather than assumed to be successful.
Capability inventory: The generated outputs facilitate shell command execution via docker and kubectl (serving-runbooks.md).
Sanitization: The skill relies on human-in-the-loop verification and explicit safety rules that prohibit the inlining of credentials like HF_TOKEN in runbooks.

omh-inference-serving