CtrlK
BlogDocsLog inGet started
Tessl Logo
SkillPublic

roboflow-training-and-evaluation

Use when training Roboflow models, diagnosing why a model underperforms, improving accuracy, or setting up a production feedback loop — covers architecture selection, model IDs, checkpoints, Model Evaluation (confusion matrix, per-class metrics, MCP eval tools), the diagnosis-first improvement playbook, and Active Learning through the Project Model Workflow block.

Quality

88%

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

81%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 dense, factual reference skill with strong workflow sequencing, explicit error-recovery guidance, and well-signaled one-level-deep references. The main improvements available are trimming the duplicated model-selection guidance and completing the MCP tool parameter signatures.

DimensionReasoningScore

Conciseness

The body is dense with platform-specific facts Claude cannot know (architecture/size/resolution tables, error codes, plan entitlements, NAS prerequisites) with almost no explanation of known concepts. It sits at the 4 anchor rather than 5 because of trimmable redundancy — the "Model Selection Quick Guide" restates the decision tree (self-acknowledged: "This summarises the training branches of the decision tree above") and Related Pages repeats links already given inline.

4 / 5

Actionability

Gives exact invocations and parameters ("trainings_create(project_id, version_number, model_type=\"rfdetr-nas-parent\")"), error codes with prescribed responses ("insufficient_validation_images_for_nas ... do not retry it"), and threshold specifics ("confidence 0-100; defaults to the optimal threshold"). It is not the 5 anchor because most rows of the MCP Tools Reference give bare tool names without signatures, and the actual model ID values are deferred to model-ids.md.

4 / 5

Workflow Clarity

The Training Flow sequence, the 14-step numbered decision tree with explicit trial/validation branches ("User confirms works → Done. Poor results → Step 13"), the NAS prerequisites checklist, error-recovery handling for named failure codes, and the cost guard ("do not start a different training until the user explicitly approves that model and its credit usage") collectively match the 5 anchor — explicit validation steps, feedback loops, and checklists. Not the 4 anchor because checkpoints are explicit rather than merely "most present".

5 / 5

Progressive Disclosure

References are one level deep and clearly signaled with purpose ("Read model-ids.md before calling a training tool. It contains the exact supported model_id values"), matching good structure. It falls short of the 5 anchor because the ~270-line body inlines full architecture, metrics, and tool tables that could be split out, and the relative-linked files (model-ids.md, nas-results.md) are not present in the bundle; it is above the 3 anchor because references are prominent and purposeful rather than buried.

4 / 5

Total

17

/

20

Passed

Description

92%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.

A strong description with an explicit trigger clause, concrete capability enumeration, and a clearly bounded Roboflow training/evaluation niche. The only weakness is modest synonym coverage for trigger terms.

DimensionReasoningScore

Specificity

Lists multiple concrete actions — "training Roboflow models", "diagnosing why a model underperforms", "improving accuracy", "setting up a production feedback loop" — plus named sub-capabilities ("architecture selection, model IDs, checkpoints", "confusion matrix, per-class metrics, MCP eval tools"). This matches the comprehensive-coverage anchor; it is above the 4 anchor because no meaningful capability of the skill is left unlisted.

5 / 5

Completeness

Explicitly answers both questions: "Use when training Roboflow models, diagnosing why a model underperforms, improving accuracy, or setting up a production feedback loop" gives concrete when-triggers, and "covers architecture selection, model IDs, checkpoints, Model Evaluation ... Active Learning" states what. Not the 4 anchor because the when-clause is explicit with specific trigger scenarios rather than merely present.

5 / 5

Trigger Term Quality

Natural user phrases like "training Roboflow models", "underperforms", "improving accuracy", and "production feedback loop" are present, matching the good-coverage anchor. It falls short of the 5 anchor because common synonyms such as "fine-tune", "model performance", or "retrain" are missing.

4 / 5

Distinctiveness Conflict Risk

The description occupies a clear niche — Roboflow model training, diagnosis, and evaluation — with triggers unlikely to fire for deployment, annotation, or generic ML skills. Minimal conflict risk, matching the 5 anchor.

5 / 5

Total

19

/

20

Passed

Validation

93%

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

CriteriaDescriptionResult

relative_links

Relative link issues: 4 missing

Warning

Total

15

/

16

Passed

Repository

roboflow/computer-vision-skills

Path

skills/roboflow-training-and-evaluation/SKILL.md

Reviewed

Table of Contents

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.