Borrowing it
Nothing to install: this file belongs to cloudrift-ai/emmy. Take a copy, put it at the same path in your own repository, and replace the rules that are about this project with yours.
curl -O https://raw.githubusercontent.com/cloudrift-ai/emmy/main/.agents/skills/onboard-model/SKILL.mdgit clone --depth 1 https://github.com/cloudrift-ai/emmyWrote this? Show the measurements
A badge with what this costs and how it scanned, read live from this page, so it follows the numbers instead of freezing them. Markdown for a README, HTML for a documentation site or a project page.
[](https://agentmods.dev/skills/cloudrift-ai/emmy/onboard-model)<a href="https://agentmods.dev/skills/cloudrift-ai/emmy/onboard-model"><img src="https://agentmods.dev/badge/skills/cloudrift-ai/emmy/onboard-model.svg" alt="Measured on agentmods" height="20"></a>- NVIDIA SkillSpector pass
What it costs to keep this loaded
Counted locally with the o200k_base tokenizer, which is exact for GPT models; Claude uses its own tokenizer and its counts differ. Treat this as one consistent yardstick across the catalogue rather than a bill. Prices are per million input tokens.
| Model | Per session | Once invoked |
|---|---|---|
| Fable 5.1 | $0.00084 | $0.07246 |
| Opus 5 | $0.00042 | $0.03623 |
| Sonnet 5 | $0.00017 | $0.01449 |
| Haiku 4.5 | $0.00008 | $0.00725 |
Grade A, and why
onboard-model scanned grade A with 0 findings against 26 rules in 11 categories — prompt injection, anti-refusal, data exfiltration, privilege escalation, supply chain, agent snooping, system-prompt leakage, SSRF and excessive agency — measured 2d ago.
A static scan of the body, not an audit. Every finding is printed with the line that produced it so you can judge whether it matters here. A mod is markdown that instructs an agent; that is exactly why what it instructs is worth reading.
Nothing flagged
None of the 26 patterns this scan looks for appear in this file: no shell pipes, no recursive deletes, no credential paths, no hidden text, no instruction-override or anti-refusal phrasing, no agent-config snooping. That is not a guarantee, it is the absence of the things that are checkable.
How it starts
The opening of the file, as written. The whole thing — 487 lines — stays where its author put it; the contents beside it link to each section on GitHub.
Onboard or reverify a model
Turn a Hugging Face model ID, an operation mode, and an exact (GPU name, GPU count) into reviewed repository
artifacts:
- one recommended serving recipe under
recipes/<model>/recipe.yaml; - one reusable serving experiment under
experiments/<model>/with a cumulative report and one Git LFS evidence archive, including the system-only row records, per exact GPU platform; - one compact, self-contained
recipes/<model>/RESULTS.mdonly when a valid final deployment recipe exists; - a complete compiler golden under
recipes/<model>/golden/when full coverage qualifies; - when Emmy is eligible, tuned kernels and a verified, prebuilt
cloudriftai/vllm-emmy-<model-slug>:<tag>image.
Repository storage retains reproducibility input and durable evidence for each qualified serving platform. In the
serving experiment root, commit recipe.yaml, one cumulative RESULTS.md, and
results_<gpu-short>x<gpu-count>.tar.gz for each measured exact GPU name/count. Store the system-only experiment
records inside that platform archive as part of the timestamped raw run; do not commit them as top-level files. Derive
<gpu-short> with emmy.hardware.gpu_short_name; for example, a single RTX 4090 uses
results_rtx4090x1.tar.gz. Track these archives
with Git LFS. When the caller says LFS is configured locally, verify the archive attribute but do not modify or list
.gitattributes; the caller owns that infrastructure file. Do not commit the ignored dated run directory, loose
benchmark JSON/TXT/logs, plots, compiler run summaries, partial working goldens, or onboarding-summary files.
Use one serving experiment root for all GPU platforms that share the protocol. On a platform-specific run, replace
only that platform's archive, update only its section of the shared experiment RESULTS.md, and preserve every other
platform archive and report section. Remove legacy top-level records only for the current platform after verifying
they are retained inside its archive. Reuse the existing serving experiment root when it already represents the
protocol; otherwise create experiments/<model>/serving/.
Use only the supplied SSH server. The caller owns VM creation and deletion; this skill owns deployed workloads and must tear them down before returning. Never switch GPU type, count, provider, model quantization, or model checkpoint to rescue a failed run.
What this file has done since we first saw it
Hashed on every crawl. A supply-chain change to an agent config is a question of when, not whether, so the history is kept rather than the latest state alone.
- 2d ago Changed · +4 lines f4fe0c2b362a
- 8d ago First seen · 483 lines · 84 tokens per session scan A 3776e960d105
onboard-model is a skill published in the GitHub repository cloudrift-ai/emmy (80 stars, last pushed today), licensed Apache-2.0. It adds 84 tokens to every session and 7,246 once invoked, about $0.0004 per session on Opus 5. A static security scan graded it A with 0 findings. No closer match exists in the catalogue, so it is treated as the original; first seen 2026-08-30.
Other skills, from other repositories
inference-aiops
Use this skill whenever the user needs to operate a GPU inference cluster — vLLM (OpenAI API + Prometheus /metrics) and Ray Serve / Ray Jobs (Ray dashboard), plus the single-process serving engines SGLang and TGI (Text Generation Inference): a one-shot cluster overview (deployments + total replicas + queue…
sglang-diffusion-performance
Use when choosing the fastest SGLang Diffusion flags for a model, GPU, and VRAM budget.
sglang-diffusion-add-model
Use when adding a new diffusion model or Diffusers pipeline to SGLang.
sglang-diffusion-modelopt-quant
Use when quantizing a diffusion DiT with NVIDIA ModelOpt and making the resulting FP8 or NVFP4 checkpoint loadable, verifiable, and benchmarkable in SGLang Diffusion.
hqq-quantization
Half-Quadratic Quantization for LLMs without calibration data. Use when quantizing models to 4/3/2-bit precision without needing calibration datasets, for fast quantization workflows, or when deploying with vLLM or HuggingFace Transformers.
together-ai-inference
Serverless inference, fine-tuning, embeddings, image generation, and batch processing on 200+ open-source models via an OpenAI-compatible API. Use when you need fast, cost-effective access to open-source LLMs without managing infrastructure.