learning-mechanics

Apply the "learning mechanics" framework — the emerging physics-style theory of deep learning training dynamics — when writing, debugging, scaling, or tuning neural-network code. Use for decisions about learning rate / batch size / width / depth scaling, μP (Maximal Update Parameterization) and hyperparameter transfer, edge-of-stability and progressive sharpening, lazy vs. rich (feature-learning) regimes and initialization scale, neural scaling laws, gradient-flow conservation laws / symmetries, neural collapse, the neural feature ansatz, and for designing scientific experiments on training. Trigger phrases: "learning mechanics", "why is training unstable", "how should I scale the learning rate with width/batch", "muP / mup / hyperparameter transfer", "edge of stability", "lazy vs rich", "feature learning regime", "scaling laws", "tune hyperparameters on a small model". Source: Simon et al., "There Will Be a Scientific Theory of Deep Learning" (2026).

Quality

92%

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

Quality

Content

85%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-engineered reference skill: dense and competent-assuming prose, concrete formulas and a symptom-routing table, a numbered workflow with prediction/validation feedback loops, and clean one-level-deep progressive disclosure into core-method.md. The only drag is mild cross-section redundancy and the absence of executable code examples (acceptable for a theory-skill, but it caps actionability at 4).

Suggestions

Collapse the 'cannot predict scaling exponents a priori' caveat so it lives in one place (e.g., decision rule #8) rather than being restated in the routing checklist and honesty notes — this would tighten conciseness toward a 5.

Add one minimal runnable snippet (e.g., logging update-to-weight ratio and a sharpness proxy, or a μP transfer pseudo-config) so the most common actions are copy-paste ready rather than formula-only.

Consider moving the 'Experiment tenets' and 'Discretization Hypothesis' framing into core-method.md to keep SKILL.md purely as the actionable overview, sharpening the overview/reference split.

Dimension	Reasoning	Score
Conciseness	The body is information-dense and assumes Claude's DL competence rather than re-explaining basics, but it deliberately repeats a few points across the decision rules, routing checklist, and honesty notes (e.g., the 'cannot predict scaling exponents a priori' caveat appears three times), keeping it just short of 'every token earns its place'.	4 / 5
Actionability	Concrete, specific guidance abounds — formulas ('η = η₀ · width^c', 'scale LR with √(batch size)'), thresholds ('λ_max ≳ 2/η'), and a symptom→rule routing table — but as an instruction-only theory skill it offers no copy-paste code/commands and some levers are conditional ('reduce the output multiplier / init scale'), matching 'mostly executable guidance with minor gaps'.	4 / 5
Workflow Clarity	A clearly ordered 6-step workflow ('do these in order — don't skip step 2') with an explicit predict-before-run checkpoint (step 4), a validate-and-update feedback loop (step 6), and supporting reproducibility/output checklists, matching the 'clear sequence with explicit validation steps; feedback loops; checklists' anchor.	5 / 5
Progressive Disclosure	SKILL.md serves as the applied-decision-rule overview while deeper rationale ('the why') is split into a single, clearly signaled, one-level-deep, verified reference file (references/core-method.md) that itself terminates without further nesting, matching the 'clear overview with well-signaled one-level-deep references' anchor.	5 / 5
	Total	18 / 20 Passed

Description

100%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.

The description is exemplary: it states concrete capabilities, gives an explicit trigger-phrase list with synonyms, answers both 'what' and 'when', and occupies a distinctive niche. Voice is imperative (consistent with the rubric's good examples) with no first/second-person pronouns to penalize.

Dimension	Reasoning	Score
Specificity	Names multiple concrete actions ('writing, debugging, scaling, or tuning neural-network code') plus a comprehensive list of specific decision domains (LR/batch/width/depth scaling, μP, edge-of-stability, lazy vs. rich, scaling laws, neural collapse), matching the 'comprehensive coverage' anchor.	5 / 5
Completeness	It explicitly answers what ('Apply the learning mechanics framework... when writing, debugging, scaling, or tuning neural-network code') and when ('Use for decisions about...' plus concrete trigger phrases), satisfying the both-what-and-when-with-trigger-phrases anchor.	5 / 5
Trigger Term Quality	An explicit 'Trigger phrases:' list gives nine natural phrases a user would say ('why is training unstable', 'how should I scale the learning rate with width/batch') including synonyms ('muP / mup'), matching the comprehensive-synonym anchor.	5 / 5
Distinctiveness Conflict Risk	It carves a clear, narrow niche (physics-style theory of DL training dynamics) with distinctive triggers that would not fire for unrelated skills, matching the 'clear niche with distinct triggers; minimal conflict risk' anchor.	5 / 5
	Total	20 / 20 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: Ares960826/ares-agent-toolkit
Path: skills/learning-mechanics/SKILL.md
Commit: d2d5ed7

Reviewed: 1 day 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.