Portfolio case study
CV Repro Lab Skills
Public ClawHub releases for benchmark-gated CV experimentation, browser validation, and promotion gating.
Overview
I turned a reproducible CV experimentation workflow into two public, installable ClawHub skills for teams running browser-heavy and GPU-heavy vision work. The releases package experiment records, browser notebook run records, heartbeat-aware VM execution, review dashboards, and promotion bundles that separate semantic, runtime, and product-surface checks.
What It Covers
- Packages benchmark-gated CV experimentation into two public ClawHub skills teams can install and reuse
- Captures reproducible experiment state with run cards, dataset manifests, review dashboards, and redacted public context snapshots
- Validates Colab, Kaggle, and browser-driven CV workflows with browser run cards and per-image validation scorecards
- Adds campaign planning and claim review with contamination checks, rerun policy, and benchmark metrics
Stack And Topics
- ClawHub
- OpenClaw Skills
- Python
- PyTorch
- Computer Vision
- Google Colab
- Kaggle
- MLOps
- Release Engineering
Public Signals
- ClawHub downloads: 1,439 total public ClawHub listings, 2026-06-04 (783 data-science-cv-repro-lab + 656 sota-agent)
- Published versions: 24 total public ClawHub listings, 2026-06-04 (12 + 12 packages)
- Live packages: 2 data-science-cv-repro-lab + sota-agent
- Execution surfaces: 3 semantic, runtime, and product-surface promotion gates
- Structured helpers: 29 scripts manifests, scorecards, summaries, and claim-review tools