pytorch-training

Best practices for building robust PyTorch training loops. Use when generating or reviewing ML training code.

Quality

73%

Does it follow best practices?

Run evals on this skill

Adds up to 20 points to the overall score

View guide

Securityby

Passed

No findings from the security scan

Fix and improve this skill with Tessl

tessl review fix ./researchclaw/skills/builtin/tooling/pytorch-training/SKILL.md

Quality

Content

72%Weight 40%Scale 1-3

Reviews the quality of instructions and guidance provided to agents. Good implementation is clear, handles edge cases, and produces reliable results.

The body is admirably lean and surfaces specific PyTorch APIs, but it reads as a tip checklist rather than an executable, sequenced workflow. It lacks a complete runnable code example and explicit validation feedback loops for batch training.

Suggestions

Add a complete, copy-paste-ready training-loop code block to the body so the skill is executable, not just a list of API pointers.

Reframe the items as a sequenced workflow with explicit validation checkpoints (e.g., validate loss/grad norms each epoch, halt on NaN, restore best checkpoint).

Surface the frontmatter code-template into the body or reference it explicitly, since the body currently has no runnable example.

Dimension	Reasoning	Score
Conciseness	The body is a tight 8-item list where every line delivers a specific API pointer with no padding or explanation of concepts Claude already knows. It is not a 2 because there is no unnecessary explanation to trim — every token earns its place.	3 / 3
Actionability	Items name concrete APIs with arguments (e.g., "DataLoader with num_workers>0 and pin_memory=True", "optimizer.zero_grad(set_to_none=True)"), but the body contains no complete, copy-paste-ready executable code block and omits key setup details. It is not a 3 because the guidance is fragmentary rather than fully executable, and not a 1 because the API references are specific rather than vague.	2 / 3
Workflow Clarity	The numbered items are best-practice tips rather than a sequenced workflow, and while early stopping and checkpointing hint at validation, there is no explicit validate→fix→retry feedback loop. Because training is a batch operation, the missing explicit validation checkpoint caps this at 2 rather than 3.	2 / 3
Progressive Disclosure	The skill is well under 50 lines with no need for external references, and the single clearly-titled section with a numbered list is well organized. Per the rubric's simple-skill guidance, this satisfies progressive disclosure with just well-organized sections.	3 / 3
	Total	10 / 12 Passed

Description

75%Weight 40%Scale 1-3

Based on the skill's description, can an agent find and select it at the right time? Clear, specific descriptions lead to better discovery.

The description clearly answers both what the skill does and when to use it, with a distinct PyTorch-training niche. It is held back from the top tier by generic action verbs and incomplete coverage of common trigger terms.

Suggestions

Replace generic verbs like "building" with specific concrete actions (e.g., "Set up reproducible training loops, configure optimizers and schedulers, add checkpointing and early stopping").

Broaden trigger coverage with natural variations users say, such as torch, deep learning, neural network, and model training.

Optionally mirror the metadata trigger keywords inside the description so the natural-language trigger matches the indexed terms.

Dimension	Reasoning	Score
Specificity	"Best practices for building robust PyTorch training loops" names a concrete domain and the action of building training loops, but it does not enumerate multiple specific actions the way a 3-anchor example does. It is not a 1 because the domain and an action are explicitly named rather than being vague like "Helps with documents".	2 / 3
Completeness	It states what the skill does ("building robust PyTorch training loops") and gives an explicit "Use when generating or reviewing ML training code" trigger, answering both what and when. It is not a 2 because the when-clause is explicit rather than merely implied.	3 / 3
Trigger Term Quality	It surfaces natural terms a user would say — "PyTorch", "training loops", "ML training code", "generating or reviewing" — but misses common variations such as torch, deep learning, neural network, or model training. Coverage is relevant but incomplete, matching the 2 anchor rather than the broad 3-anchor term list.	2 / 3
Distinctiveness Conflict Risk	"PyTorch training loops" is a clear, narrow niche with distinct triggers unlikely to fire for unrelated skills. It is not a 2 because the PyTorch + training-loop framing is specific enough to avoid overlap with generic document or data skills.	3 / 3
	Total	10 / 12 Passed

Validation

100%

Warnings & errors only

Checks the skill against the spec for correct structure and formatting. All validation checks must pass before discovery and implementation can be scored.

Validation — 16 / 16 Passed

Validation for skill structure

No warnings or errors.

Repository: aiming-lab/AutoResearchClaw
Path: researchclaw/skills/builtin/tooling/pytorch-training/SKILL.md
Commit: e2e23c9

Reviewed: 6 days ago

Table of Contents

Discovery Implementation Validation

Is this your skill?

If you maintain this skill, you can claim it as your own. Once claimed, you can manage eval scenarios, bundle related skills, attach documentation or rules, and ensure cross-agent compatibility.