tbvjvsladla

tbvjvsladla/easy_vllm_simulator

Skill Claude CodeCodex

A performance-testing guide for an active vLLM server, which serves machine-learning models and generates their responses.

tbvjvsladla/easy_vllm_simulator

Skill Claude CodeCodex

A tool for publishing, indexing, and distributing verified model-serving recipes using hint tags. A serving recipe is a set of settings or steps for running a model, while indexing makes those recipes easier to find.

tbvjvsladla/easy_vllm_simulator

Skill Claude CodeCodex

An onboarding coordinator for a fresh clone or an environment that has not yet been configured. It first determines whether the setup uses one machine or multiple machines, then scans the relevant nodes and records them in a shared manifest.yaml file.

tbvjvsladla/easy_vllm_simulator

Skill Claude CodeCodex

A watcher for upstream vLLM, the open-source system used to serve language models. It detects newer versions when asked, explains the changes, and suggests updates to the project's version layers.

tbvjvsladla/easy_vllm_simulator

Skill Claude CodeCodex

A recipe planner for serving one fixed AI model with vLLM, a system for running language models as a service. It reads the model's config.json, estimates graphics-memory use for different settings, filters out options that exceed the budget or safety margin, ranks the rest, and can create three serving files for DGX Spark.

tbvjvsladla/easy_vllm_simulator

Skill Claude CodeCodex

Path-reference librarian over the project's docs/ work-history (devlog, testlog, plan, simlog) and the sub-doc mirror. Builds a deterministic relationship graph (cites/realizes/evidences edges) and surfaces authority-ranked, edge-traversable context WITHOUT copying source bodies. Use when starting an Ouroboros…

tbvjvsladla/easy_vllm_simulator

Instructions file

A set of Claude Code instructions for the easy_vllm_simulator project. The instructions describe the project's goals, rules that must remain true, safety boundaries, and actions that should trigger specific guidance.

tbvjvsladla/hermes-subagent-control-plane

Skill Claude CodeCodex

Use when an orchestrator must control a real Claude Code CLI sub-agent in another workspace. Reads the target agent-card.json, runs an auth/HOME preflight (no hardcoded paths), proves workspace context with a read-only discovery call, delegates bounded claude -p work with the correct workdir, and verifies results…