Decision-focused SKILL.md files for AI coding agents working on computer vision projects.
This repository provides opinionated decision frameworks, common gotchas, and pipeline recipes for image processing tasks. Unlike generic OpenCV tutorials, these skills focus on when to use what, in what order, and what traps to avoid.
AI coding agents already know OpenCV's API. What they lack is:
- Decision guidance: "Which filter for this noise type?"
- Pipeline ordering: "What MUST come before Canny edge detection?"
- Parameter gotchas: "
blockSizemust be odd,resize()takes (x,y) not (y,x)" - Domain-specific rules: "Solidity ≈ 1.0 = benign, < 1.0 = malign in tumor analysis"
| # | Skill | Focus | Level |
|---|---|---|---|
| 1 | Image Fundamentals | Color spaces, formats, coordinate traps | 🟢 Beginner |
| 2 | Preprocessing Decisions | Filter selection tree, noise identification | 🟡 Intermediate |
| 3 | Thresholding Strategy | Global vs Otsu vs Adaptive decision matrix | 🟡 Intermediate |
| 4 | Morphology Toolkit | Opening vs Closing, structuring element selection | 🟡 Intermediate |
| 5 | Contour Analysis | Shape metrics, medical classification ratios | 🔴 Advanced |
| 6 | YOLO Pipeline | Detection, segmentation, classification, XAI | 🔴 Advanced |
| 7 | MediaPipe Tracking | Face, hand, pose landmark tracking | 🔴 Advanced |
Ready-to-use processing chains:
| Recipe | Use Case |
|---|---|
| Medical X-Ray Preprocessing | CLAHE → Filter → Threshold → Morph |
| Object Counting | Threshold → Morph → Contour → Count |
| Color Object Tracking | HSV → inRange → Mask → Bitwise |
| YOLO Training Checklist | Data → Label → Augment → Train → Evaluate |
Minimal, runnable Python scripts demonstrating each skill:
| Script | Demonstrates |
|---|---|
| filter_comparison.py | Mean vs Gaussian vs Median vs Bilateral |
| threshold_comparison.py | Global vs Otsu vs Adaptive |
| morphology_demo.py | Erosion, Dilation, Opening, Closing |
| contour_features.py | Area, perimeter, solidity, centroid |
| hsv_color_filter.py | HSV masking pipeline |
| yolo_inference.py | YOLO detect/segment/classify |
| mediapipe_pose.py | MediaPipe pose estimation |
This repo follows the SKILL.md standard. To use with your AI coding agent:
- Clone this repo into your project or reference it
- AGENTS.md at root provides the global context
- Each SKILL.md is self-contained and can be loaded independently
- Pipeline recipes provide copy-paste ready processing chains
Compatible with: Cursor, Claude Code, GitHub Copilot, Gemini CLI, and any agent that reads .md context files.
MIT — Use freely in any project.
Built by Ali Eren — Backend .NET Developer & AI Engineering enthusiast at Pamukkale University.