Engineering algorithms, deterministic evaluation harnesses, and multi-agent context systems — bridging precision metrology (Python / Rust / C++ / FPGA) with verifiable AI workflows. Based in Shanghai (Xuhui · Caohejing).
| Repository | Focus |
|---|---|
| swe-rl-sandbox | Verifiable SWE-RL environment & defense-in-depth anti-reward-hacking grader (multi-stage Docker isolation, AST + HMAC completion guards, metrology & exchange-ledger tasks). |
| agent-memory-layers | Zero-dependency dual-axis engineering memory control plane (standalone & multi-host hub), zero-token mechanical validator (aml check), and 5-arm ablation rubric scorer (aml eval). |
| optics-device-survey | Evidence-graded hardware due-diligence skill for coding agents: first-principles threshold derivation, condition normalization, and hard-gate-before-score rubrics. |
| eng-coach | Cross-platform UserPromptSubmit lifecycle hook and /eng skill installer for Claude Code with idempotent settings merge. |
| coprime-determinant-count | Combinatorial derivation and computational verification of |
- storytold/wordcraft#7441 (Rust · 18k+ ★) — Cross-crate inline IME composition layout reservation & reflow across
wordcraft-layout,wordcraft-engine, andwordcraft-ui-egui(CJK–Latin boundary autospace suppression, non-mutatingSessionpreview preserving undo/Track Changes/a11y invariants, and multi-crate unit & canvas tests).