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 pengzhangzhi/superpowers-ml --skill ml-experiment-designgit clone --depth 1 https://github.com/pengzhangzhi/superpowers-mlWrote 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/pengzhangzhi/superpowers-ml/ml-experiment-design)<a href="https://agentmods.dev/skills/pengzhangzhi/superpowers-ml/ml-experiment-design"><img src="https://agentmods.dev/badge/skills/pengzhangzhi/superpowers-ml/ml-experiment-design/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/pengzhangzhi/superpowers-ml/ml-experiment-design"><img src="https://agentmods.dev/badge/skills/pengzhangzhi/superpowers-ml/ml-experiment-design.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 | $0.00042 | $0.00767 |
| Opus 5 | $0.00021 | $0.00383 |
| Sonnet 5 | $0.00008 | $0.00153 |
| Haiku 4.5 | $0.00004 | $0.00077 |
Grade A, and why
ml-experiment-design 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 11d 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 — 62 lines — stays where its author put it; the contents beside it link to each section on GitHub.
ML Experiment Design
Overview
An ML idea is not runnable until it is a CARD. This skill turns a rough idea into a compact, one-screen experiment card — rung R0, the precondition for planning or launching anything. Fill it WITH your human partner; do not write training code until the card is settled.
Core principle: One change under test, one locked metric, one decision rule. If you cannot state in advance what result would change your mind, you are not ready to run.
Upstream: the idea comes from superpowers-ml:brainstorming.
Downstream: the card drives superpowers-ml:ml-feedback-ladder (how to verify it cheaply) and superpowers-ml:writing-plans (the tasks).
The Experiment Card
Settle every field. Keep each to a line or two — a card, not a document.
- Research question — the one question this run answers.
- Hypothesis — specific and falsifiable: what changes, and the direction you expect.
- Baseline — the exact, runnable comparison (config / commit, not "the usual setup").
- Variant(s) — the SINGLE change under test. Anything else is a separate experiment.
- Primary metric — one metric, locked now, that decides the outcome.
- Guardrail metrics — what must NOT regress while the primary metric moves (cost, latency, memory, a quality/safety metric).
- Dataset / split — train / val / test, and the leakage risk you checked for.
- Seed policy — how many seeds, fixed or swept; how you tell seed noise from a real effect.
- Budget — the compute / wall-clock ceiling you will spend before stopping to decide.
- Cheapest useful rung — the smallest rung that tells you something real, to start at (see
superpowers-ml:ml-feedback-ladder). - Success criterion — the threshold on the primary metric that CONFIRMS the hypothesis, set before running.
- Exploratory-only — what this run may suggest but can never confirm (anything the locked primary metric does not measure).
One Change At A Time
A variant is ONE change, or you cannot attribute the result — two changes is two experiments. If your human partner wants to vary many things, that is a study: write a card per change, then design a ladder for each.
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.
- 11d ago First seen · 62 lines · 42 tokens per session scan A 33ca42749e9c
ml-experiment-design is a skill published in the GitHub repository pengzhangzhi/superpowers-ml (8 stars, last pushed 3d ago), licensed MIT. It adds 42 tokens to every session and 767 once invoked, about $0.0002 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 2026-08-31.
Other skills, from other repositories
hugging-face-model-trainer
This skill should be used when users want to train or fine-tune language models using TRL (Transformer Reinforcement Learning) on Hugging Face Jobs infrastructure. Covers SFT, DPO, GRPO and reward modeling training methods, plus GGUF conversion for local deployment. Includes guidance on the TRL Jobs package, UV…
hugging-face-datasets
Create and manage datasets on Hugging Face Hub. Supports initializing repos, defining configs/system prompts, streaming row updates, and SQL-based dataset querying/transformation. Designed to work alongside HF MCP server for comprehensive dataset workflows.
marimo-pair
Work inside the user's live marimo notebook from the code editor: run Python in the same kernel the user does, inspect live notebook state, and commit durable notebook changes through code mode. Use whenever you create, analyze, or improve the user's marimo notebook.
grpo-finetune
Fine-tune a model with GRPO on Fireworks-managed GPUs from a plain-English task description and a dataset. Use this skill whenever the user wants to fine-tune, RL-tune, or GRPO-train a model on their own data — or says things like "train a model to extract/classify/score X", "fine-tune on this dataset", "set up a GRPO…
training-check
Interactively monitor training metrics from the current Codex session, periodically checking WandB or fallback logs for NaN, divergence, plateaus, and broken runs.
physicsnemo-shard-tensor
Official NVIDIA-authored guidance for PhysicsNeMo ShardTensor domain parallelism — integrate domain parallelism into training/inference scripts (new or existing) with DDP or FSDP2, write and register shard patches to enable new layers/ops, and bootstrap multi-GPU correctness tests. Use when working with ShardTensor…