aris-training-check

Periodically check WandB metrics during training to catch problems early (NaN, loss divergence, idle GPUs). Avoids wasting GPU hours on broken runs. Use when training is running and you want automated health checks.

Quality

85%

Does it follow best practices?

Run evals on this skill

Adds up to 20 points to the overall score

View guide

Securityby

Medium

Suggest reviewing before use

Quality

Content

77%

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

A highly actionable, clearly sequenced workflow with strong code examples and validation checkpoints. It loses points for redundant watchdog framing and a dangling external reference with no progressive-disclosure bundle structure.

Suggestions

Consolidate the watchdog distinction into a single section to remove the repetition between 'When to Use' and 'Integration with Watchdog'.

Move the detailed Codex prompt template and signal/judgment tables into a references file (e.g., references/judgment.md) and link from SKILL.md to implement one-level-deep progressive disclosure.

Verify the '../../tools/watchdog.py' path resolves from the skill's install location, or describe watchdog's role inline without relying on a link that may not exist.

Dimension	Reasoning	Score
Conciseness	The body is mostly efficient and does not lecture on concepts Claude knows, but it restates the description in the intro and repeats the watchdog-vs-training-check distinction in both 'When to Use' and the 'Integration with Watchdog' table, which could be consolidated.	2 / 3
Actionability	Provides fully executable WandB API code, a concrete SSH log-fallback command, and a complete Codex MCP prompt with exact config and a constrained STOP/CONTINUE/WAIT response format — copy-paste ready.	3 / 3
Workflow Clarity	A clear four-step sequence (Read → Judgment → Codex → Act) with explicit decision tables, a WandB-unreachable fallback, escalation only when unsure, and an evidence-saving step on the destructive 'Stop' action, satisfying the explicit-validation-checkpoint anchor.	3 / 3
Progressive Disclosure	No bundle files exist (references/, scripts/, assets/ absent) and the only external pointer — '../../tools/watchdog.py' — is not present in the workspace; the skill is a single well-sectioned file but offers no one-level-deep detailed references for the sizable Codex/judgment material.	2 / 3
	Total	10 / 12 Passed

Description

92%

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 strong, concrete description with explicit trigger guidance and natural training-domain keywords. Its only weakness is minor overlap with process-health monitoring via the 'idle GPUs'/'automated health checks' phrasing, which the body itself reassigns to a separate tool.

Suggestions

Drop 'idle GPUs' from the description (it is process-health, not training-quality) to sharpen the niche and avoid overlap with watchdog.py.

Tighten 'automated health checks' to training-specific language (e.g., 'automated training-quality checks') to reduce conflict with process-health skills.

Dimension	Reasoning	Score
Specificity	Lists multiple concrete checks — 'check WandB metrics during training to catch problems early (NaN, loss divergence, idle GPUs)' and 'Avoids wasting GPU hours on broken runs' — matching the 'lists multiple specific concrete actions' anchor.	3 / 3
Completeness	Explicitly answers both what ('Periodically check WandB metrics... to catch problems early') and when ('Use when training is running and you want automated health checks'), satisfying the explicit-trigger anchor.	3 / 3
Trigger Term Quality	Covers natural terms a user would say when training a model — 'WandB metrics', 'training', 'NaN', 'loss divergence', 'idle GPUs', 'GPU hours' — giving good coverage of common phrasings.	3 / 3
Distinctiveness Conflict Risk	The niche is fairly specific, but 'idle GPUs' and 'automated health checks' overlap the process-health watchdog domain that the skill body explicitly assigns to watchdog.py, creating some overlap with a sibling skill rather than a fully distinct niche.	2 / 3
	Total	11 / 12 Passed

Validation

81%

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 — 13 / 16 Passed

Validation for skill structure

Criteria	Description	Result
allowed_tools_field	'allowed-tools' contains unusual tool name(s)	Warning
frontmatter_unknown_keys	Unknown frontmatter key(s) found; consider removing or moving to metadata	Warning
relative_links	Relative link issues: 2 suspicious	Warning

	Total	13 / 16 Passed

Repository: OpenLAIR/dr-claw
Path: skills/aris-training-check/SKILL.md
Commit: 6da7f7c

Reviewed: about 7 hours 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.