ruview-model-training

Train RuView models — camera-free WiFlow pose (10 sensor signals, no labels), camera-supervised pose (MediaPipe + ESP32 CSI → 92.9% PCK@20, ADR-079), RuVector contrastive embeddings (AETHER, ADR-024), domain generalization (MERIDIAN, ADR-027), local SNN environment adaptation, plus GPU training on GCloud and Hugging Face publishing. Use when building, fine-tuning, evaluating, or shipping a model.

Quality

75%

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 ./plugins/ruview/skills/ruview-model-training/SKILL.md

Quality

Content

65%

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

A dense, highly actionable skill body organized into clear tracks with mostly executable commands. Its weaknesses are inline time-sensitive metrics that will stale, validation that is decoupled from the per-track and publishing workflows, and reference material that should be split into bundled files but is not.

Suggestions

Move volatile details (version numbers, benchmark timings, GPU cost/hr, GCloud project IDs) into a separate 'versions / costs' reference file or a deprecated-style section so the main body stays evergreen and token-lean.

Integrate a validation checkpoint into the publishing workflow (e.g. run the test/verify suite and require VERDICT: PASS before 'publish-huggingface') and add an explicit 'if validation fails, fix and re-run' recovery loop, since publishing is an outward-facing irreversible action.

Pull the inline ADR list, crate list, and data-layout reference into bundled reference files under references/ and link to them one level deep, so SKILL.md remains a true overview.

Dimension	Reasoning	Score
Conciseness	The body is mostly lean (terse prose plus executable commands, no explaining of concepts Claude knows), but it is peppered with time-sensitive ephemeral detail — 'RuVector v2.0.4', '~84 s on an M4 Pro', '92.9% PCK@20', project 'cognitum-20260110', and per-GPU cost figures — that is not isolated in a deprecated/old-patterns section and will go stale, so it cannot score the lean-anchor 3.	2 / 3
Actionability	Nearly every section gives real, copy-paste-ready commands ('cargo run -p wifi-densepose-sensing-server -- --train ...', 'python scripts/collect-ground-truth.py', 'gcloud ...', 'bash scripts/gcloud-train.sh --gpu l4'), matching the fully-executable anchor despite minor placeholders like '--epochs <N>'.	3 / 3
Workflow Clarity	Track B provides a clear numbered collect/train/evaluate sequence and there is a dedicated validation section with expected outcomes ('1,400+ pass, 0 fail', 'VERDICT: PASS'), but validation is a detached post-change step rather than an integrated per-track checkpoint, the Hugging Face publishing workflow has no pre-publish verification gate, and no explicit 'if-fail-fix-retry' recovery loop is given — fitting the 'validation gaps' anchor.	2 / 3
Progressive Disclosure	References are clearly signaled and one level deep (e.g. 'See docs/adr/ADR-079-camera-ground-truth-training.md'), but no bundle files actually exist in references/scripts/assets to offload detail into, and substantial reference material (the inline ADR list, crate list, data-layout table) lives in SKILL.md rather than being split out, so content is not 'appropriately split across files'.	2 / 3
	Total	9 / 12 Passed

Description

85%

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 highly specific, third-person description with an explicit 'Use when' trigger and a clear niche. Its main weakness is trigger-term quality: natural user phrasing is buried under a heavy load of internal project codenames and metric jargon.

Suggestions

Lead with the natural user-facing trigger ('train, fine-tune, evaluate, or publish RuView models') and push internal codenames (AETHER, MERIDIAN, ADR-079, PCK@20) into the body so the description reads in terms a user would actually say.

Add common trigger variations users might phrase (e.g. 'WiFi pose estimation', 'contrastive CSI embeddings', 'model retraining') to broaden natural-keyword coverage.

Dimension	Reasoning	Score
Specificity	Lists multiple concrete actions — 'camera-free WiFlow pose', 'camera-supervised pose', 'RuVector contrastive embeddings', 'domain generalization', 'local SNN environment adaptation', 'GPU training on GCloud and Hugging Face publishing' — far exceeding the anchor that merely names a domain and some actions.	3 / 3
Completeness	Clearly answers both what (the enumerated training tracks) and when via the explicit 'Use when building, fine-tuning, evaluating, or shipping a model' trigger, so it is not capped at 2 by a missing 'Use when' clause.	3 / 3
Trigger Term Quality	The 'Use when building, fine-tuning, evaluating, or shipping a model' clause supplies natural user terms, but the description is dominated by internal codenames (AETHER, MERIDIAN, ADR-079, PCK@20, RaBitQ) a user would never say, so coverage of natural terms is partial rather than comprehensive.	2 / 3
Distinctiveness Conflict Risk	The RuView-specific niche and project codenames make it clearly distinguishable and unlikely to trigger for an unrelated skill; it is well above the 'somewhat specific but could overlap' anchor.	3 / 3
	Total	11 / 12 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
referenced_paths_exist	Referenced path issues: 18 missing	Warning

	Total	15 / 16 Passed

Repository: ruvnet/RuView
Path: plugins/ruview/skills/ruview-model-training/SKILL.md
Commit: d9dfea2

Reviewed: about 13 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.