CtrlK
BlogDocsLog inGet started
Tessl Logo
SkillPublic

pytorch-lightning

Deep learning framework (PyTorch Lightning / lightning package). Organize PyTorch code into LightningModules, configure Trainers for multi-GPU/TPU, implement data pipelines, callbacks, logging (W&B, TensorBoard, MLflow), distributed training (DDP, FSDP, DeepSpeed), for scalable neural network training.

Quality

83%

Does it follow best practices?

Run evals on this skill

Adds up to 20 points to the overall score

View guide
SecuritybySnyk

Passed

No findings from the security scan

SKILL.md
Quality
Evals
Security

Quality

Content

86%Weight 40%Scale 1-5

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

A well-structured, highly actionable skill body: an executable smoke test, pinned version guidance, and clean one-level-deep disclosure into verified reference and script files. Its weaknesses are minor — a few generic passages that could be trimmed and a Quick Workflow lacking an explicit error-recovery loop.

Suggestions

Trim the generic overview paragraph and the "When to Use This Skill" list (which restates the frontmatter description), and remove the duplicate CSV entry in the logging section.

Add a brief failure-recovery note to the Quick Workflow (e.g., what to check if the smoke test errors) to upgrade the workflow from a single checkpoint to a validate-fix-retry loop.

DimensionReasoningScore

Conciseness

The body is efficient and dense with version-specific, non-obvious guidance (NeptuneLogger removal tied to the service shutdown date, sync_dist=True default semantics, DDP duplicate padded samples) and pushes detail into references. Minor trimmable instances remain: the generic "PyTorch Lightning is a deep learning framework" overview, a "When to Use This Skill" list that duplicates the frontmatter description, generic Trainer feature bullets, and CSV appearing twice in the logging list. This fits anchor 4 (efficient, minor over-explanation) better than anchor 3 (some unnecessary explanation).

4 / 5

Actionability

Guidance is fully executable: a pinned install command, concrete strategy configuration (Trainer(strategy="ddp", accelerator="gpu", devices=4)), a complete copy-paste CPU smoke test in the Quick Workflow, and pointers to three executable script templates. Specific examples cover the common train/validate/test case, matching the anchor-5 example.

5 / 5

Workflow Clarity

The Quick Workflow gives a clear sequence (copy scripts, run the smoke test, then follow real-data guidance) and the smoke run itself acts as a validation checkpoint, with explicit notes on what it does and does not validate ("fast_dev_run disables much of that behavior"). It falls short of anchor 5 because there is no error-recovery feedback loop or checklist for failure cases.

4 / 5

Progressive Disclosure

The body is a genuine overview: seven component sections each signal exactly one one-level-deep reference (references/trainer.md, references/logging.md, etc.) or executable script, all of which exist in the bundle, and a Resources section lists every script and reference with one-line descriptions. This matches the anchor-5 pattern of a concise getting-started body with well-signaled references.

5 / 5

Total

18

/

20

Passed

Description

71%Weight 40%Scale 1-5

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

A specific, tool-rich description that comprehensively states what the skill does, weakened by the absence of any explicit 'Use when' trigger guidance and by missing high-frequency natural terms like checkpointing. It is clearly distinguishable from most skills but has minor overlap risk with generic PyTorch/deep-learning skills.

Suggestions

Add an explicit trigger clause, e.g. "Use when the user mentions PyTorch Lightning, lightning, LightningModule, Trainer, or asks about distributed training (DDP, FSDP, DeepSpeed) or experiment logging (W&B, TensorBoard, MLflow)."

Include high-frequency natural terms users actually say, such as checkpointing, ModelCheckpoint, early stopping, and mixed precision, to round out trigger coverage.

Tighten the opening so it leads with the named framework rather than the generic phrase "Deep learning framework", reducing conflict risk with a general PyTorch skill.

DimensionReasoningScore

Specificity

The description lists six concrete action areas with named tools ("Organize PyTorch code into LightningModules, configure Trainers for multi-GPU/TPU, implement data pipelines, callbacks, logging (W&B, TensorBoard, MLflow), distributed training (DDP, FSDP, DeepSpeed)"), comprehensively covering the framework's surface. It exceeds the anchor-4 example because coverage spans the whole capability set rather than leaving gaps.

5 / 5

Completeness

The 'what' is clearly and concretely stated, but there is no 'Use when...' clause or equivalent explicit trigger guidance — only the weakly implied purpose "for scalable neural network training". Per the rubric guideline, a missing explicit trigger clause caps completeness at 3, which is the exact anchor match.

3 / 5

Trigger Term Quality

Good natural-term coverage including synonyms and package names ("PyTorch Lightning / lightning package", "LightningModules", "DDP, FSDP, DeepSpeed", "W&B, TensorBoard, MLflow"), but common user phrases like checkpointing, mixed precision, and fine-tuning are missing. This fits anchor 4 (good coverage, a few natural terms missing) better than the comprehensive anchor 5.

4 / 5

Distinctiveness Conflict Risk

The named framework and Lightning-specific terms (LightningModule, DDP/FSDP/DeepSpeed, lightning package) create a clear niche, but the generic opening "Deep learning framework... Organize PyTorch code" creates minor overlap risk with a general PyTorch or deep-learning skill. Anchor 4 (mostly distinct, minor overlap with closely related skills) is the best fit.

4 / 5

Total

16

/

20

Passed

Validation

100%

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

K-Dense-AI/scientific-agent-skills

Path

skills/pytorch-lightning/SKILL.md

Reviewed

Table of Contents

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.