mixed-precision

Use FP16/BF16 mixed precision to accelerate training and reduce memory. Use when optimizing GPU performance.

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/mixed-precision/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.

A concise, well-organized body that assumes Claude's competence and names specific APIs. It would benefit from a complete executable example and a more explicit sequenced workflow with validation checkpoints.

Suggestions

Add a short, complete executable code block in the body (e.g. autocast + GradScaler training loop) so the guidance is copy-paste ready rather than directive-only.

Reorder the bullets into a clear sequenced workflow (wrap forward in autocast -> scale loss -> step -> update) with an explicit validation checkpoint for NaN gradients instead of a passive watch note.

Clarify the BF16-vs-FP16 selection as a concrete decision step (check GPU arch, then choose dtype) rather than a single preference bullet.

Dimension	Reasoning	Score
Conciseness	The body is a lean bullet list with no padding and no explanation of concepts Claude already knows; every token earns its place, matching the score-3 anchor.	3 / 3
Actionability	Bullets name specific APIs ('torch.cuda.amp.autocast()', 'GradScaler') giving concrete direction, but the body lacks a complete, copy-paste executable code block, fitting the score-2 anchor of concrete-but-incomplete guidance.	2 / 3
Workflow Clarity	The content is a bullet list of best practices rather than a clearly sequenced workflow, and the only checkpoint ('Watch for NaN gradients — reduce learning rate if needed') is implicit rather than an explicit validation step, matching the score-2 anchor.	2 / 3
Progressive Disclosure	The skill is under 50 lines with no external references needed, and its single well-organized section suffices, matching the simple-skill allowance for a score-3 on progressive disclosure.	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.

A concise, third-person description that clearly answers both what and when with a distinct technical niche. Its main weakness is a narrow trigger clause and limited enumeration of concrete actions.

Dimension	Reasoning	Score
Specificity	The description names the domain ('FP16/BF16 mixed precision') and two concrete effects ('accelerate training and reduce memory'), but does not enumerate a comprehensive set of specific actions, matching the score-2 anchor rather than the multiple-distinct-actions bar of 3.	2 / 3
Completeness	It states both what the skill does ('Use FP16/BF16 mixed precision to accelerate training and reduce memory') and when to use it ('Use when optimizing GPU performance') with an explicit trigger, matching the score-3 anchor.	3 / 3
Trigger Term Quality	It surfaces relevant natural terms (fp16, bf16, mixed precision, training, GPU, memory), but the explicit trigger clause 'Use when optimizing GPU performance' is narrow and omits common phrasings like 'training is slow' or 'out of memory', fitting the score-2 anchor of some keywords missing common variations.	2 / 3
Distinctiveness Conflict Risk	Mixed-precision training is a clear, specialized niche with distinct triggers (fp16/bf16, mixed precision) unlikely to collide with other skills, matching the score-3 anchor.	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/mixed-precision/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.