Getting it into your agent
One page per mod, every tool's command on it. A separate URL per tool would split the same page into five that compete with each other.
npx agentmods add skills/cloudrift-ai/emmy/release-serving-imagenpx skills add cloudrift-ai/emmy --skill release-serving-imagegit 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/release-serving-image)<a href="https://agentmods.dev/skills/cloudrift-ai/emmy/release-serving-image"><img src="https://agentmods.dev/badge/skills/cloudrift-ai/emmy/release-serving-image.svg" alt="Measured on agentmods" height="20"></a>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 | $0.00094 | $0.04801 |
| Opus 5 | $0.00047 | $0.02400 |
| Sonnet 5 | $0.00019 | $0.00960 |
| Haiku 4.5 | $0.00009 | $0.00480 |
Grade C, and why
release-serving-image scanned grade C with 1 finding 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 today.
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.
Recursive force deletehighDestructive command
rm -rf with a variable or a broad path is one typo away from removing the wrong tree.
`rm -rf venv` before the bake — nothing after the warm needs it. How it starts
The opening of the file, as written. The whole thing — 283 lines — stays where its author put it; the contents beside it link to each section on GitHub.
Release a prebuilt serving image for one model
The steps, their rationale, and the cache-key parity contract live in
docker/vllm-emmy-serve/ARCHITECTURE.md — read it first; this
skill does not restate it. What the skill adds is orchestration: model and mode selection, detached execution
that survives SSH drops, hard PASS/FAIL gates that abort-and-teardown instead of rationalizing past a bad signal,
the headroom-sweep policy that finalizes the model config, secret hygiene for the push, and the release
side-effects.
The recipe is the publication input. Resolve exactly one model.huggingface, one concrete engine image, one hardware
target, and one serving shape before provisioning; reject experiment grids or multiple image variants. Use the
resolved HF id as MODEL for the build and qualification commands:
make serve-config MODEL=google/gemma-4-12B-it # model / slug / config path / image tag / target GPU
make serve-models # which models already have a pinned config
The recipe image must use the canonical immutable reference that emmy publish validates:
cloudriftai/<runtime-family>-<model-slug>:<runtime-version>-<source-sha>
Supported runtime families are vllm-emmy and 1cat-vllm. The model slug comes only from model.huggingface
through Emmy's slug implementation. GPU architecture, zero-recompile status, cache manifests, checkpoint revision,
and serving shape are image labels and qualification evidence, never tag suffixes. Refuse latest, jitfree, GPU
names, or mutable aliases in the primary recipe reference.
The exact local image must label ai.emmy.publish.family, ai.emmy.model.id, ai.emmy.model.revision,
ai.emmy.target.gpu, org.opencontainers.image.version, and org.opencontainers.image.revision; cache-bearing
images also carry their cache-manifest digests. Compare the model, checkpoint revision, target GPU, sealed serving
shape, and cache manifests against the recipe, model config, and qualification evidence before publication. A label
does not substitute for the fresh-container zero-recompile gate.
Budget: ~2–3 h wall on a rental for a 12B-class model (image build ~20 min, headroom sweep ~30–60 min, validate ~30 min, warm ~30 min, bake/verify/publication ~30 min). Hard cap: 4 h — if the session exceeds it, capture logs, tear down, and report. A larger checkpoint scales this up; re-estimate before starting rather than inheriting these numbers.
What ships with it
1 file beside SKILL.md in the same directory: the scripts, references and assets a skill reads on demand. Not counted in the per-session cost; read them before you install if any of them is executable.
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.
- today Changed · +4 lines a3b9bf5dcda6
- 5d ago First seen · 279 lines · 94 tokens per session scan C 5e520e40fd90
release-serving-image is a skill published in the GitHub repository cloudrift-ai/emmy (80 stars, last pushed today), licensed Apache-2.0. It adds 94 tokens to every session and 4,801 once invoked, about $0.0005 per session on Opus 5. A static security scan graded it C with 1 finding (recursive force delete). 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…
huawei-cloud-ascend-models-deploy
Huawei Cloud Ascend model deployment and testing skill for large language models on Ascend DevServer (910B series). Supports single-machine and dual-machine deployment for LLM, VL (vision-language), Embedding, and Rerank models. Provides model inference testing, deployment log viewing, and status monitoring with…
astrea
Use for hipfire quant calibration, imatrix-driven experiments, KLD/PPL quality evaluation, k-map/format selection, MQ/HFQ/HFP/MFP tradeoff work, ParoQuant-style weight transform planning, and KV policy planning. Use when deciding whether a calibrated model candidate should be promoted, rejected, packaged, or sent…
lambda-labs-gpu-cloud
Reserved and on-demand GPU cloud instances for ML training and inference. Use when you need dedicated GPU instances with simple SSH access, persistent filesystems, or high-performance multi-node clusters for large-scale training.
lambda-labs-gpu-cloud
Reserved and on-demand GPU cloud instances for ML training and inference. Use when you need dedicated GPU instances with simple SSH access, persistent filesystems, or high-performance multi-node clusters for large-scale training.
lambda-labs-gpu-cloud
Reserved and on-demand GPU cloud instances for ML training and inference. Use when you need dedicated GPU instances with simple SSH access, persistent filesystems, or high-performance multi-node clusters for large-scale training.