senior-computer-vision

Took a look at your computer vision skill — the domain coverage is solid, but the spec could use some tightening to really guide users through the complexity here. What's the core problem you're trying to solve with this one?

Links:

• GitHub
• Your agentic skill in the SkillzWave marketplace

The TL;DR

You're at 45/100, firmly in F territory. This is graded against Anthropic's best practices for agentic skills. Your strongest area is Spec Compliance (12/15) — the frontmatter and naming conventions are solid. But Utility (6/20) is dragging you down hard. The skill reads generic instead of vision-specific, and the reference files are basically identical boilerplate.

What's Working Well

• Valid YAML structure — Your frontmatter is clean and follows conventions properly
• Consistent naming — senior-computer-vision uses proper hyphen-case formatting
• Grep-friendly structure — The skill has decent header organization for searchability

That said, these are table-stakes stuff. You need the substance to back them up.

The Big One: Zero Computer Vision Content

Here's the core problem: you've built a computer vision skill that contains almost zero actual computer vision content. You've got 227 lines talking about "world-class senior professionals" and generic MLOps, but nothing about CNNs, object detection models, segmentation architectures, or vision-specific optimization techniques.

Why this matters: When someone invokes this skill, they're asking for help with vision problems — not data engineering platitudes. The skill should guide them on YOLO architectures, Transformer-based detection (DETR), segmentation models (Mask R-CNN, SAM), video analysis, 3D vision. Right now it's indistinguishable from a generic ML skill.

The fix: Rewrite the core sections with actual vision content. Reference computer_vision_architectures.md should cover CNN vs ViT trade-offs, object detection architectures, when to use YOLO vs Faster R-CNN vs DETR. object_detection_optimization.md should dive into NMS variants, anchor optimization, loss function choices. production_vision_systems.md should cover ONNX/TensorRT deployment, batch inference optimization, edge device considerations. Include specific libraries: torchvision, detectron2, mmdetection, ultralytics.

Impact: +10 points

Other Things Worth Fixing

1. Duplicate reference files are wasting tokens — All three reference files are identical boilerplate. Split them into genuinely different content (architectures, optimization, production) instead of copy-paste. (+8 points)

2. Placeholder commands don't exist — Lines 14-23 reference scripts/vision_model_trainer.py that doesn't exist. Either create actual working scripts or swap in real examples using standard libraries like yolo detect train or python -m mmdet.apis test. (+5 points)

3. "World-class" appears 5+ times — Kills your objectivity score. Replace marketing fluff with specific claims: "Achieves >95% mAP on COCO detection benchmarks" instead of "world-class capabilities". (+4 points)

4. Missing table of contents — 227 lines needs a TOC for navigation. Add one after frontmatter with anchors to main sections. (+2 points)

Quick Wins

Most impactful first:

• Strip marketing language and replace with vision-specific technical depth (+10 points)
• Differentiate your three reference files with actual content branches (+8 points)
• Replace placeholder scripts with real, runnable commands (+5 points)
• Add marketing objectivity fixes and TOC (+6 points)

That's +29 points reachable with focused rewrites. You'd land around 74/100 — solid B-range territory.

Checkout your skill here: SkillzWave.ai | SpillWave We have an agentic skill installer that install skills in 14+ coding agent platforms. Check out this guide on how to improve your agentic skills.