Skill Claude CodeCodex
Populate or refresh the autotune DB's cross-hardware node table with measurements from a specific GPU. Use when the user asks to collect node data on an existing or rented GPU, run the budgeted three-slice golden sweep, merge the GPU-keyed node rows into the local tuning DB, back up the DB, and tear down any server…
Skill Claude CodeCodex
Use this skill when the user asks what new models to try or benchmark, wants newly released open models discovered, wants trending models mapped to suitable GPU hardware, or wants the maintained recipe set refreshed. It produces a ranked shortlist or lifecycle selection ready for the onboard-model skill, using keyless…
Skill Claude CodeCodex
Onboard or periodically reverify and benchmark a Hugging Face model on an exact target GPU platform. Use when asked to add a model recipe, refresh a maintained recipe on a supplied GPU server, benchmark serving, create reproducible experiments and a durable results report, fully qualify and tune the model's Emmy…
Skill Claude CodeCodex
Use this skill when the user asks to release, build, warm, bake, qualify, or publish a per-model serving image, including requests to run emmy publish RECIPE. Runs the documented vllm-emmy release workflow on the model's target GPU, covering golden realization and memory gates, HF parity, warm and bake, offline…
Skill Claude CodeCodex
Use this skill when the user asks to re-run an article's benchmarks, reproduce blog post numbers, validate that an article URL still holds, check whether the latest code still performs like a published post, or otherwise compare re-measured Emmy results with published results. It fetches the article, finds its…
Skill Claude CodeCodex
Run or rerun Emmy experiment recipes, including requests to adjust an experiment harness before running it, then preserve per-platform compressed raw results with system-only YAML experiment records and a thoughtful cumulative RESULTS.md interpretation. Use for requests such as "run this experiment", "benchmark this…
Skill Claude CodeCodex
Provision a fresh cloud GPU VM without immediately deploying a model, then leave it available for later benchmarks or inference. Use when the user asks to start a remote server, spin up a GPU VM, provision cloud GPU capacity, or create a server with a specified GPU type and count.
Skill Claude CodeCodex
Tune Emmy kernels for a Hugging Face model, traced IR, or golden YAML. Use when asked to tune a model or golden set, seed MCTS with model-proposed knob configurations, compare hybrid proposals against MCTS-only search, diagnose slow or failing kernels, refresh per-GPU goldens, or produce a per-kernel tuning findings…
Agent Claude Code
Senior Architect reviewer. Reviews plans and diffs for simplicity, duplication, encapsulation and abstraction. Read-only — never edits code. Use before opening a PR, or when a design decision needs a second opinion.
Agent
Size one candidate checkpoint onto exact GPU deployments from its published configuration.
Agent
Find current open-weight model momentum and exact identities on Hugging Face.
Agent
Refresh the Emmy model lifecycle through bounded, read-only research.
Agent
Find current served models and quality signals through OpenRouter and LMArena.
Agent
Find recently discussed open models on Reddit as an independent discovery source.
Agent
Score one exact batch of existing recipes from shared discovery evidence.
Agent
Investigate one bounded model-onboarding compatibility or failure question.
Agent
Qualify one model on the workflow-owned GPU and produce reviewed Emmy artifacts.
Instructions file CodexOpenCode
Instructions for cloudrift-ai/emmy, covering agents.md, project overview, prerequisites, running tests and the realization corpus.
Instructions file
Instructions for cloudrift-ai/emmy, a project described as: Optimized GPU compiler for LLM inference. Choose from a list of optimized recipes or optimize your own model via kernel fusion, autotuning, and advanced scheduling. Run benchmarks across different GPU types and configurations, track results and share…
Command
Command "ARCHITECTURE" from cloudrift-ai/emmy, covering commands architecture, layered design, layers, emmy/recipe/ — recipe library and emmy/deploy/ — deploy library.