Tessl • The management layer for AI agent skills

SkillPublic

pytorch-patterns

PyTorch deep learning patterns and best practices for building robust, efficient, and reproducible training pipelines, model architectures, and data loading.

Quality

72%

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 ./.kiro/skills/pytorch-patterns/SKILL.md

The canonical home for this skill is pytorch-patterns in affaan-m/ECC

Quality

Content

71%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-organized, highly actionable single-file PyTorch pattern library whose main weaknesses are length-driven redundancy and no progressive-disclosure split. The good/bad contrast style and quick-reference table are effective, but several sections overlap and no external references partition the material.

Suggestions

Deduplicate content that appears in multiple sections: keep mixed-precision guidance in one place (the Performance section's AMP example largely restates the training-loop code) and fold the eval-mode anti-pattern into the Validation Loop section it duplicates.

Split the pattern groups into one-level-deep reference files (e.g., references/training.md, references/data-pipelines.md, references/performance.md) and keep SKILL.md as a concise overview with clearly signaled links, reducing inline token cost.

Trim patterns that restate standard PyTorch knowledge Claude already has (basic Dataset/DataLoader usage) and retain only the non-obvious guidance (set_to_none=True, weights_only=True, use_reentrant=False, drop_last for BatchNorm).

Dimension	Reasoning	Score
Conciseness	The body is dense and code-heavy, but at ~390 lines it repeats itself (mixed precision appears in both 'Training Loop Patterns' and 'Performance Optimization'; eval-mode/no_grad guidance recurs in 'Anti-Patterns') and much of it (standard training loops, DataLoader flags, Dataset classes) restates knowledge Claude already has. It fits 'mostly efficient but includes some unnecessary explanation or could be tightened' rather than 2, since there is no conceptual padding or beginner-level prose.	3 / 5
Actionability	Every section provides complete, executable, copy-paste-ready code (train_one_epoch with AMP/GradScaler, evaluate, ImageDataset, DataLoader config, save/load checkpoint, torch.compile), and the quick-reference table covers the common idioms. This matches the top anchor: 'Fully executable; copy-paste ready code or commands; specific examples cover the common cases.'	5 / 5
Workflow Clarity	Sections are clearly organized by task, and each pattern (train epoch, checkpoint round-trip) is internally unambiguous with correct ordering (zero_grad -> forward -> backward -> step). Not 5: no validation or feedback checkpoints are shown (e.g., verifying a checkpoint loads, or a profile-then-fix loop), so it sits below 'explicit validation steps; feedback loops for error recovery'.	4 / 5
Progressive Disclosure	No bundle files exist, and the single SKILL.md holds the entire ~390-line pattern library with good section headers but no split into reference files. This matches 'some structure but could be better organized; content that should be separate is inline' — the training/data/performance pattern groups are natural candidates for separate reference files — rather than 4, which would require most content to be appropriately placed in separate files.	3 / 5
	Total	15 / 20 Passed

Description

61%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 domain-clear, reasonably keyword-rich description that identifies the PyTorch niche well but lacks any 'when to use' trigger clause and pads the what with vague qualifiers. Its main gaps are the missing trigger guidance and thin concrete action coverage.

Suggestions

Add an explicit trigger clause, e.g., 'Use when writing or reviewing PyTorch training code, model architectures, or data loading pipelines' — without it, completeness is capped at 3.

Replace vague qualifiers ('robust, efficient, and reproducible') with concrete actions users would recognize, such as 'Implements training loops, AMP mixed precision, checkpointing, and reproducible seeding'.

Include natural synonyms users actually say — 'neural network training', 'GPU/CUDA', 'training script', '.pt checkpoints' — to round out trigger-term coverage.

Dimension	Reasoning	Score
Specificity	The description names the domain ('PyTorch deep learning patterns and best practices') with one concrete action ('building') applied to three targets ('training pipelines, model architectures, and data loading'), but 'patterns and best practices' plus qualifiers 'robust, efficient, and reproducible' are generic filler. This matches 'names domain and 1-2 concrete actions, but not comprehensive'; it is not 4 because it does not list several distinct specific actions.	3 / 5
Completeness	The 'what' is clear (PyTorch patterns for pipelines, architectures, data loading) but there is no 'Use when...' clause or equivalent trigger guidance anywhere in the description, which per the judging guidelines caps completeness at 3 ('has a clear what but when is missing or only weakly implied').	3 / 5
Trigger Term Quality	Strong natural terms users would say: 'PyTorch', 'deep learning', 'training pipelines', 'model architectures', 'data loading'. A few common variations are missing (e.g., 'neural network', 'GPU', 'training scripts', file extensions like .pt), matching 'good keyword coverage; a few natural terms missing' rather than the comprehensive synonym-plus-extension coverage of 5.	4 / 5
Distinctiveness Conflict Risk	'PyTorch' pins a clear niche that most skills would not conflict with, but the broad 'deep learning patterns' framing leaves minor overlap risk with general ML/deep-learning or Python-performance skills, and no explicit trigger phrase separates it. This matches 'mostly distinct; minor overlap risk with closely related skills' rather than 5's 'clear niche with distinct triggers'.	4 / 5
	Total	14 / 20 Passed

Validation

93%

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 — 15 / 16 Passed

Validation for skill structure

Criteria	Description	Result
frontmatter_unknown_keys	Unknown frontmatter key(s) found; consider removing or moving to metadata	Warning

	Total	15 / 16 Passed

Repository: affaan-m/ECC
Path: .kiro/skills/pytorch-patterns/SKILL.md
Revision: b38be6a

Reviewed: 2 days ago

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.

pytorch-patterns

Table of Contents

Is this your skill?