CtrlK
BlogDocsLog inGet started
Tessl Logo
SkillPublic

tao-train-pose-classification

Pose classification using ST-GCN (Spatial Temporal Graph Convolutional Network). Classifies skeleton sequences into action categories from pose-keypoint data. Use when training, evaluating, exporting, or running inference for a TAO pose-classification model. Trigger phrases include "train pose classification", "skeleton action recognition", "ST-GCN", "keypoint sequence classifier".

Quality

85%

Does it follow best practices?

Run evals on this skill

Adds up to 20 points to the overall score

View guide
SecuritybySnyk

Passed

No findings from the security scan

SKILL.md
Quality
Evals
Security

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.

The content is highly actionable with concrete, executable docker commands and per-action spec overrides, and its conditional routing and error patterns give it good workflow clarity. Its weaknesses are repeated instructions that inflate token cost and references to a schemas/ directory that does not exist in the bundle, which undermines navigation.

Suggestions

State the 'skip dataset_convert when .npy/.pkl files are provided' rule once (e.g. in the Dataset Convert section) and reference it from the other sections instead of repeating it in the intro, Typical Spec Overrides, and Error Patterns.

Package the referenced schemas/<action>.schema.json and schemas/manifest.json, or correct the Dataclass Schemas and Train Action Policy sections so every referenced path exists in the bundle.

Remove the duplicated '--shm-size=8g' docker flag and consolidate the val_loss vs accuracy guidance currently split between Training Requirements and Eval Dataset.

DimensionReasoningScore

Conciseness

The body is dense and free of concept padding, but the dataset_convert skip rule is repeated four times (intro, Typical Spec Overrides, Dataset Convert section, Error Patterns), the val_loss vs accuracy guidance appears twice, and the docker common block duplicates '--shm-size=8g'. Mostly efficient but noticeably could be tightened.

3 / 5

Actionability

Copy-paste-ready docker run commands for all five actions, complete spec_overrides JSON blocks per action, exact artifact filenames (e.g. 'model_epoch_000_step_00007.pth'), and concrete error patterns with specific fixes fully cover the common cases.

5 / 5

Workflow Clarity

Conditional routing is explicit (dataset_convert only for raw BodyPose JSON; AutoML on/off policy with a missing-schema fallback) and the checkpoint-handoff and accuracy-verification guidance act as checkpoints, but there is no explicit numbered end-to-end sequence tying the actions together with per-step validation.

4 / 5

Progressive Disclosure

Sections are well organized and the references/ bundle files are one level deep, but the body repeatedly cites 'schemas/<action>.schema.json' and 'schemas/manifest.json' which are not packaged in the bundle, and the spec templates are only mentioned generically rather than linked per-section, leaving structure and signaling incomplete.

3 / 5

Total

15

/

20

Passed

Description

95%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 a strong, well-structured example: clear third-person 'what', an explicit 'Use when' clause, and a comprehensive set of natural trigger phrases covering both lay and technical terminology. The only nitpick is that supported operations are listed inside the trigger clause rather than the capability statement.

DimensionReasoningScore

Specificity

The 'what' statement gives one concrete action ('Classifies skeleton sequences into action categories from pose-keypoint data') while train/evaluate/export/inference are enumerated only inside the 'when' clause rather than as stated capabilities, leaving minor gaps versus anchor 5.

4 / 5

Completeness

It explicitly answers both what ('Classifies skeleton sequences into action categories from pose-keypoint data') and when ('Use when training, evaluating, exporting, or running inference for a TAO pose-classification model') with concrete trigger phrases, matching the anchor-5 example pattern exactly.

5 / 5

Trigger Term Quality

Explicit trigger phrases ('train pose classification', 'skeleton action recognition', 'ST-GCN', 'keypoint sequence classifier') combine natural user phrasing with technical synonyms, comprehensively covering the domain; there are no canonical file extensions for skeleton data to add.

5 / 5

Distinctiveness Conflict Risk

'TAO pose-classification model', 'ST-GCN', and 'keypoint sequence classifier' carve a clear niche with minimal conflict risk; the pose-classification qualifier distinguishes it from sibling TAO model skills and from pose estimation.

5 / 5

Total

19

/

20

Passed

Validation

87%

Checks the skill against the spec for correct structure and formatting. All validation checks must pass before discovery and implementation can be scored.

Validation — 14 / 16 Passed

Validation for skill structure

CriteriaDescriptionResult

frontmatter_unknown_keys

Unknown frontmatter key(s) found; consider removing or moving to metadata

Warning

referenced_paths_exist

Referenced path issues: 2 missing

Warning

Total

14

/

16

Passed

Repository

NVIDIA/skills

Path

skills/tao-train-pose-classification/SKILL.md

Revision

fce3cca

Reviewed

Table of Contents

DiscoveryImplementationValidation

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.

Claim skill