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 mayank-io/mstack --skill ml-design-review-iterativegit clone --depth 1 https://github.com/mayank-io/mstackWrote 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/mayank-io/mstack/ml-design-review-iterative)<a href="https://agentmods.dev/skills/mayank-io/mstack/ml-design-review-iterative"><img src="https://agentmods.dev/badge/skills/mayank-io/mstack/ml-design-review-iterative/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/mayank-io/mstack/ml-design-review-iterative"><img src="https://agentmods.dev/badge/skills/mayank-io/mstack/ml-design-review-iterative.svg" alt="Reviewed on agentmods" width="80" 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 5.1 | $0.00062 | $0.01545 |
| Opus 5.5 | $0.00025 | $0.00618 |
| Sonnet 5.5 | $0.00012 | $0.00309 |
| Haiku 4.5 | $0.00006 | $0.00154 |
Grade A, and why
ml-design-review-iterative 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 2mo 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 — 103 lines — stays where its author put it; the contents beside it link to each section on GitHub.
Iterative ML Design Review
Run up to 4 review-fix iterations with 3 parallel reviewer agents (Principal ML Engineer, Sr Applied Scientist, ML Ops Engineer). Each iteration raises the bar for what gets fixed, converging quickly to a statistically sound and practically deployable design.
Review Scope
By default, review the most recently written or edited design doc in docs/. If the user specifies a file, use that instead.
Process
Step 0: Initialize
Set iteration = 1. Identify the design doc to review. Read it fully before launching reviewers. Also read CLAUDE.md and any referenced prior designs or experiment logs to understand the project context.
Step 1: Launch 3 Reviewer Agents in Parallel
Dispatch 3 agents simultaneously, each reviewing the same design doc from a different perspective. Every agent MUST:
- Research relevant frameworks, papers, or tools mentioned in the design (use web search)
- Categorize each finding as exactly one of: BLOCKER, HIGH, MEDIUM, LOW, NIT
- Include a confidence score (0-100) for each finding
Agent 1 — Principal ML Engineer:
Review this design document as a Principal ML Engineer with deep experience in AutoML, neural architecture search, and automated research systems. Focus on: evaluation methodology soundness, metric design (can the optimization target be Goodharted?), overfitting risks given the dataset size, data leakage vectors, search space design, whether the evaluation is deterministic and reproducible, whether baselines are sufficient to validate the harness, and whether the system will actually find real signal vs. noise. Research any referenced frameworks (e.g., autoresearch, optuna, ray tune) to verify the design faithfully adapts their principles. Categorize every finding as BLOCKER, HIGH, MEDIUM, LOW, or NIT with a confidence score.
Stay in your lane: You own ML methodology, evaluation design, and search architecture. Do NOT rule on domain-specific assumptions about the data (e.g., whether a financial feature is meaningful) — that is the Applied Scientist's domain. Do NOT rule on infrastructure choices — that is ML Ops' domain. If you see a concern outside your lane, flag it as a question for the relevant reviewer.
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.
- 2mo ago First seen · 103 lines · 62 tokens per session scan A 0cb2e2c580f6
ml-design-review-iterative is a skill published in the GitHub repository mayank-io/mstack (6 stars, last pushed 3d ago), licensed MIT. It adds 62 tokens to every session and 1,545 once invoked, about $0.0002 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 2026-08-31.
Other skills, from other repositories
huggingface-llm-trainer
Train or fine-tune language models with TRL or Unsloth on Hugging Face Jobs, including SFT, DPO, GRPO, reward models, and GGUF conversion. Use for cloud LLM training; use huggingface-vision-trainer for vision tasks.
huggingface-vision-trainer
Train object-detection, image-classification, or SAM segmentation models on Hugging Face Jobs. Use for vision fine-tuning and evaluation; use huggingface-llm-trainer for language models.
submit-github-pr
Use when publishing an existing TensorRT-Model-Connect change as a GitHub pull request. Verifies authenticated repository access, branch and diff scope, validation evidence, commit identity, reviewer-facing text, exact pushed head, and the created draft PR without merging it.
huggingface-best
Find and compare recommended Hugging Face models for a task using benchmarks, model size, and device constraints. Use for model selection questions; use huggingface-local-models for GGUF setup and hf-cli for Hub operations.
council-review
Multi-model validation council — auto-validate plans, architecture changes, and PRs via validate-plan/review before executing.
ring:adopting-lib-commons-huma-wrapper
Adopting the lib-commons/v5 shared Huma (OAS 3.1) OpenAPI wrapper + RFC 9457 problem model (commons/net/http/{openapi,problem}) in a Lerian Go service: wire openapi.New/ServeSpec + problem.Install (central >=500 scrub) on BOTH runtime and spec-gen paths, the per-rail problem.MapError flex seam, and rename-only spec…