diepre-vision-cognitionDiePre 视觉认知 Skill —— 将包装/模切机器视觉感知与 SOUL 推理融合的认知框架
Install via ClawdBot CLI:
clawdbot install kingofzhao/diepre-vision-cognitionGrade Limited — based on market validation, documentation quality, package completeness, maintenance status, and authenticity signals.
Calls external URL not in known-safe list
https://github.com/KingOfZhao/AGI_PROJECTUses known external API (expected, informational)
arxiv.orgAudited Apr 17, 2026 · audit v1.0
Generated Oct 16, 2026
Manufacturers of corrugated and folding carton packaging use the skill to analyze dieline images captured on the production line, extracting deterministic features and flagging ambiguous regions for four-direction visual collision analysis. Defects falling below 90% confidence are automatically escalated to human review, ensuring no faulty die-cut reaches the customer.
Converters with decades of physical die drawings can photograph them and let the VLM-powered pipeline generate DXF/CAD code with iterative verification. The two-dimensional-to-parametric-three-dimensional path corrects perspective distortion and produces reusable digital tooling assets.
A quality team deploys the skill on inspection stations to generate an initial AI verdict on each batch, then a human reviewer confirms or corrects the label. Vision log entries written to vision_log/YYYY-MM-DD.jsonl accumulate into a continuous fine-tuning dataset that progressively improves model accuracy on site-specific materials.
Using the four-direction collision log, operators analyze side-lit captures to reveal crease cracks, surface scratches, and material thickness anomalies invisible under normal illumination. Context parameters such as material type and thickness_mm tailor confidence thresholds per substrate.
Deploy the low-latency, memory-optimized VLA-inspired architecture locally on edge hardware next to the die-cutter. This avoids cloud round-trips and keeps high-speed packaging lines running while still producing confidence-scored defect lists in real time.
Offer a cloud-hosted version of the cognition framework with dashboard, vision log storage, and confidence-threshold configuration on a monthly per-inspector seat basis. Packaging plants pay for predictable access while the vendor retains the model improvement flywheel from aggregated logs.
Sell an annual on-premises deployment license to large converters who cannot send images off-site, bundled with integration into existing MES or ERP systems and custom calibration for their die types and materials. Professional services cover initial setup, threshold tuning, and operator training.
Expose vision.analyze() as a metered API where customers pay per analyzed image or per CAD-generation request, ideal for smaller shops and third-party software vendors embedding the capability into their own packaging tools. Volume discounts and confidence-tier pricing align cost with value delivered.
💬 Integration Tip
Wrap the vision.analyze() call in your existing MES or inspection-station workflow and persist results to vision_log/ so human corrections become a continuous training signal; start with confidence threshold at 90% and tune it per material context (e.g. corrugated, thickness_mm) to balance automation rate against manual review load.
Scored Apr 19, 2026
Control desktop applications on Windows — launch, close, focus, resize, move windows, simulate keyboard/mouse input, manage processes, control VSCode, read clipboard, and capture screen info. Use when the user wants to interact with any running program, switch windows, type text, press shortcuts, open files in VSCode, manage running processes, or get system display information.
Conduct rigorous, adversarial code reviews with zero tolerance for mediocrity. Use when users ask to "critically review" my code or a PR, "critique my code", "find issues in my code", or "what's wrong with this code". Identifies security holes, lazy patterns, edge case failures, and bad practices across Python, R, JavaScript/TypeScript, SQL, and front-end code. Scrutinizes error handling, type safety, performance, accessibility, and code quality. Provides structured feedback with severity tiers (Blocking, Required, Suggestions) and specific, actionable recommendations.
Pragmatic coding standards for writing clean, maintainable code — naming, functions, structure, anti-patterns, and pre-edit safety checks. Use when writing new code, refactoring existing code, reviewing code quality, or establishing coding standards.
Claude Code integration for OpenClaw. This skill provides interfaces to: - Query Claude Code documentation from https://code.claude.com/docs - Manage subagents and coding tasks - Execute AI-assisted coding workflows - Access best practices and common workflows Use this skill when users want to: - Get help with coding tasks - Query Claude Code documentation - Manage AI-assisted development workflows - Execute complex programming tasks
Plan, draft, version, and refine written content with enforced versioning and quality audits.
Use when writing tests, creating test strategies, or building automation frameworks. Invoke for unit tests, integration tests, E2E, coverage analysis, performance testing, security testing.