The Agent Skills Directory

[COMMAND_EXECUTION]: The skill makes extensive use of local shell scripts and system commands to manage inference servers and environment detection.
The scripts/deploy.sh script automates server lifecycle management (start, stop, status) and handles GPU discovery via nvidia-smi.
It utilizes tools like ss, lsof, and ps to monitor port usage and process health.
[EXTERNAL_DOWNLOADS]: The skill fetches documentation, configuration recipes, and software packages from well-known technology providers and official repositories.
Downloads inference recipes and cookbooks from the official GitHub repositories of vLLM and SGLang.
Installs benchmarking and deployment frameworks (vllm, sglang, tensorrt-llm, aiperf) via standard package managers.
References official Docker images from NVIDIA NGC and Docker Hub for containerized deployments.
[REMOTE_CODE_EXECUTION]: The skill employs a pattern of piping network request outputs into a Python interpreter for data processing.
Evidence: In scripts/deploy.sh and references/benchmarking.md, responses from curl calls to http://localhost:8000/v1/models and /v1/chat/completions are piped to python3 -c for JSON parsing and output validation.
Context: These operations target the local inference endpoint just initialized by the user and are used exclusively for automated health checks and coherence gating during benchmarking.
[DATA_EXFILTRATION]: The skill accesses local cluster configuration files to facilitate remote deployments via SSH or SLURM.
Evidence: Accesses paths such as ~/.config/modelopt/clusters.yaml and .agents/clusters.yaml.
Context: This access is scoped to identifying authorized remote environments for deployment and does not involve exfiltrating data to external domains.

deployment