Stars
📄Score Centering Stabilizes Off-policy Reinforcement Learning
My learning notes for ML SYS.
DeepSeek Harness: Everything is a Plugin.
SQuTR- a benchmark for Spoken Query to Text Retrieval. Accepted to SIGIR'2026
你想蒸馏的下一个员工，何必是同事。蒸馏任何人的思维方式——心智模型、决策启发式、表达DNA。Distill how anyone thinks.
🔥[KDD’26] Official repository for the paper “FDABench: A Benchmark for Data Agents on Analytical Queries over Heterogeneous Data”.
A curated collection of papers and resources on On-Policy Distillation for Large Language Models.
PhoneHarness runtime harness for mixed-action phone agents
[EMNLP 2026] MobileGym: A Verifiable and Highly Parallel Simulation Platform for Mobile GUI Agent Research · 浏览器里运行的安卓模拟器 · Browser-hosted Android Simulator · Verifiable Evaluation · Scalable Onlin…
Agent skill that turns natural language, code, Terraform/K8s, SQL, OpenAPI, AsyncAPI, Protobuf and GraphQL sources into editable, tested draw.io architecture diagrams: incremental sync, multi-view …
Benchmarking Autonomous Mobile Agents in Agent-User Interactive and MCP-Augmented Environments (ACL 2026)
AndroidWorld is an environment and benchmark for autonomous agents
HTML PPT Studio — AgentSkill with 24 themes, 31 layouts, 20+ animations for building professional HTML presentations
TRACE: Trajectory Recovery for Continuous Mechanism Evolution in Causal Representation Learning
21 writing rules for AI coding and writing agents. Drop-in for Claude Code, Codex, Copilot, Cursor, and Aider, so their output reads like a tech pro.
Rethinking On-Policy Distillation of Large Language Models: Phenomenology, Mechanism, and Recipe
CoEvoSkills: Self-Evolving Agent Skills via Co-Evolutionary Verification — COLM 2026
The code used to train and run inference with the ColVision models, e.g. ColPali, ColQwen2, and ColSmol.
SkillRouter: Retrieve-and-Rerank Skill Selection for LLM Agents at Scale
InterruptBench: Evaluating LLM Agents Under User Interruptions in Long-Horizon Web Tasks
ARIS ⚔️ (Auto-Research-In-Sleep) — Lightweight Markdown-only skills for autonomous ML research: cross-model review loops, idea discovery, and experiment automation. No framework, no lock-in — works…
An offline-first scientific writing workspace powered by Claude. LaTeX + Python + 100+ scientific skills all running locally.
Skill package for ML/CV/NLP paper writing, curated and adapted from Prof. Peng Sida's open notes for Codex, Claude Code, and Gemini.
Claude Autoresearch Skill — Autonomous goal-directed iteration for Claude Code. Inspired by Karpathy's autoresearch. Modify → Verify → Keep/Discard → Repeat forever.