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 skills add metaspartan/cybara --skill huggingface-vision-trainergit clone --depth 1 https://github.com/metaspartan/cybaraWrote 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/metaspartan/cybara/huggingface-vision-trainer)<a href="https://agentmods.dev/skills/metaspartan/cybara/huggingface-vision-trainer"><img src="https://agentmods.dev/badge/skills/metaspartan/cybara/huggingface-vision-trainer/github.svg" alt="Measured on agentmods" height="20"></a>Or the 80×15 button, for a site that already has a row of RSS and ATOM ones. Only the verdict fits; the numbers stay here.
<a href="https://agentmods.dev/skills/metaspartan/cybara/huggingface-vision-trainer"><img src="https://agentmods.dev/badge/skills/metaspartan/cybara/huggingface-vision-trainer.svg" alt="Reviewed on agentmods" width="80" 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 | $0.00037 | $0.00335 |
| Opus 5.5 | $0.00015 | $0.00134 |
| Sonnet 5.5 | $0.00007 | $0.00067 |
| Haiku 4.5 | $0.00004 | $0.00034 |
Grade A, and why
huggingface-vision-trainer 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 1mo 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.
What it actually says
Hugging Face Vision Trainer
Use this workflow for image classification, object detection, segmentation, depth estimation, and vision-language fine-tuning.
Workflow
- Inspect the dataset card, license, image features, labels, splits, corrupt records, class balance, dimensions, and annotation coordinate format.
- Visualize a small representative sample before transforming labels or augmenting images.
- Choose a model and processor that explicitly support the task and license requirements.
- Preserve label mappings, preprocessing settings, image normalization, and augmentation seeds in the training configuration.
- Split by source or subject when random row splitting could leak near-duplicate images.
- Select task-appropriate metrics such as accuracy/F1, mAP, IoU, or VQA-style scores.
- Run a small batch through preprocessing, forward pass, loss, and metric computation before launching remote training.
- Save the processor with the model and publish representative evaluation outputs in the model card.
For vision-language models, ensure image tokens are not truncated and verify the model's chat template. Use PEFT when it materially reduces memory without invalidating the task.
Remote GPU work follows the huggingface-jobs confirmation and monitoring workflow. Do not upload private images, biometric data, or sensitive annotations without explicit authorization.
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.
- 1mo ago First seen · 25 lines · 37 tokens per session scan A 4133a601176f
huggingface-vision-trainer is a skill published in the GitHub repository metaspartan/cybara (32 stars, last pushed today), licensed MIT. It adds 37 tokens to every session and 335 once invoked, about $0.0001 per session on Opus 5.5. 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
baseline-comparison-review
Use when a new or more complex ML model is proposed and its value over simple baselines is not yet demonstrated -- before approving a new architecture or replacing an existing heuristic.
dataset-split-review
Use when reviewing how data is split into train/validation/test sets -- especially with time-series data, repeated entities (users, patients, sessions), or imbalanced targets.
experiment-tracking-review
Use when reviewing ML training code to confirm a run could be reconstructed later -- hyperparameters, metrics, data references, and artifacts logged, not just printed to stdout.
reproducible-training-runs
Use when reviewing or modifying ML training scripts that must produce identical results across runs or machines -- runs with the "same" config differ, or a past result must be reconstructed exactly.
Jupyter Live Kernel
Guides notebook-first analysis with reproducible kernels, inspectable data loading, and explicit promotion paths back into durable code.
seo-ai-seo
When the user wants to optimize content for AI search engines, get cited by LLMs, or appear in AI-generated answers. Also use when the user mentions 'AI SEO,' 'AEO,' 'GEO,' 'LLMO,' 'answer engine optimization,' 'generative engine optimization,' 'LLM optimization,' 'AI Overviews,' 'optimize for ChatGPT,' 'optimize for…