forgecad-reconstruction-benchmark

What problem does it solve?

This skill helps you reconstruct a visible CAD reference into readable, fully editable ForgeCAD source within a fixed benchmark submission path.

Core Features & Use Cases

• Benchmark-aware reconstruction loop: Guides a coarse-to-fine rebuild process using provided reference evidence and local scoring feedback.
• Strict submission compliance: Enforces non-negotiables like a fixed output location and bans asset-embedding shortcuts (e.g., import/read/embed-style solutions).
• Budget and grader driven iteration: Prioritizes a runnable first candidate, then iteratively improves geometry fidelity using the local RL grader and score outputs.
• Use Case: When given a reference CAD asset for an RL reconstruction episode, rebuild it as parametric ForgeCAD source at submission/main.forge.js (or task-specified path) and maximize the grader score.

Quick Start

Run the reconstruction workflow in the prepared workspace by rebuilding the reference evidence into the required submission file using the local wrappers and the RL grader as your main reward signal.