senior-computer-vision

Computer vision engineering skill for object detection, image segmentation, and visual AI systems. Covers CNN and Vision Transformer architectures, YOLO/Faster R-CNN/DETR detection, Mask R-CNN/SAM segmentation, and production deployment with ONNX/TensorRT. Includes PyTorch, torchvision, Ultralytics, Detectron2, and MMDetection frameworks. Use when building detection pipelines, training custom models, optimizing inference, or deploying vision systems.

Quality

88%

Does it follow best practices?

Run evals on this skill

Adds up to 20 points to the overall score

View guide

Securityby

—

The risk profile of this skill

Quality

Content

76%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 highly actionable, well-structured skill body with copy-paste commands and real bundle files backing every reference. Its main weaknesses are minor redundancy with the description, a CNN-vs-ViT table re-teaching known concepts, and a lack of explicit validation in the destructive dataset-cleaning workflow.

Suggestions

Add an explicit validation/dry-run checkpoint to Workflow 3 Step 2 before deleting images (e.g., a 'review the audit report, then confirm' step or a --dry-run flag) and a post-clean verification command so the destructive batch operation has a real feedback loop.

Trim the 'Core Expertise' and 'Tech Stack' sections, which restate frameworks already listed in the description, or merge them into a single concise table to reduce token overhead.

Consolidate the 'Reference Documentation' and 'Resources' sections into one set of reference pointers to remove the duplicated listings and the redundant reference-docs-and-commands.md index layer.

Dimension	Reasoning	Score
Conciseness	Mostly efficient — dense tables and copy-paste commands rather than prose — but the 'Core Expertise' and 'Tech Stack' sections restate framework lists already in the description, and the 'CNN vs Vision Transformer Trade-offs' table restates concepts Claude already knows (global context, positional encoding). Not 3 because the bulk is actionable and not padded; not 5 because those redundancies could be trimmed.	4 / 5
Actionability	Fully executable guidance throughout — concrete commands like 'python scripts/vision_model_trainer.py data/coco/ --task detection --arch yolov8m --epochs 100', 'trtexec --onnx=model.onnx --saveEngine=model.engine --fp16', and inline verification snippets ('python -c "import onnx; ...onnx.checker.check_model(model)"'); referenced scripts exist as real files in scripts/.	5 / 5
Workflow Clarity	Three workflows are each sequenced into 6 numbered steps, but Workflow 3 performs a destructive batch operation ('--remove-corrupted --remove-duplicates') with no explicit validation/verification checkpoint after deletion (the step is named 'Clean and Validate' yet shows no validation command), so per the destructive/batch cap workflow clarity cannot exceed 3. Not 2 because the audit-then-clean sequence and ONNX/benchmark verifications in other workflows do provide some checkpoints.	3 / 5
Progressive Disclosure	Good structure with a clear overview pointing to real one-level-deep reference files (computer_vision_architectures.md, object_detection_optimization.md, production_vision_systems.md, all verified present) plus scripts/. Not 5 because the 'Reference Documentation' and 'Resources' sections duplicate the same reference listings, and reference-docs-and-commands.md acts as a mild second-level index to files already reachable from the body.	4 / 5
	Total	16 / 20 Passed

Description

100%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, specific description that covers both capabilities and explicit use-when triggers in third-person voice, with comprehensive natural trigger terms for the CV domain. No significant gaps; it would be hard to misroute a request given these triggers.

Dimension	Reasoning	Score
Specificity	Names the domain plus multiple concrete actions in the 'Use when' clause — 'building detection pipelines, training custom models, optimizing inference, or deploying vision systems' — with comprehensive coverage of architectures and frameworks (YOLO/Faster R-CNN/DETR, Mask R-CNN/SAM, ONNX/TensorRT, PyTorch/Detectron2/MMDetection).	5 / 5
Completeness	Explicitly answers both 'what' (computer vision engineering for detection, segmentation, and deployment covering specific architectures/frameworks) and 'when' ('Use when building detection pipelines, training custom models, optimizing inference, or deploying vision systems') with concrete trigger phrases.	5 / 5
Trigger Term Quality	Uses natural CV vocabulary a user would actually say — 'object detection', 'image segmentation', 'YOLO', 'Faster R-CNN', 'training custom models', 'optimizing inference', 'deploying vision systems' — with comprehensive coverage of the domain's standard synonyms; not below 5 because this domain has no relevant file-extension triggers to omit.	5 / 5
Distinctiveness Conflict Risk	Occupies a clear niche (production computer vision / detection / segmentation / deployment) with distinctive triggers; unlikely to fire for unrelated skills, and uses third-person voice throughout with no first/second-person phrasing.	5 / 5
	Total	20 / 20 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: alirezarezvani/claude-skills
Path: engineering-team/skills/senior-computer-vision/SKILL.md
Commit: 19392f7

Reviewed: ~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.