Havuz

Open decision models from 0.8B to 27B: a calibrated probability for every option in one pass, and a short thought only when unsure. Engine f

NeuroArxiv, Claude Code ve Codex CLI için bir skill/CLI olarak arXiv makalelerini tarayıp teknik tasarım önerisini kaynaklara dayandırmaya çalışır.

Neden yükseliyor · 53 günde 433 yıldız alması ilgi gördüğünü gösteriyor; arXiv araştırmasını agent iş akışına eklemesi ve README'deki karşılaştırmalı eval iddiası dikkat çekmiş olabilir. Ancak eval, README'de beş problemle sınırlı ve bağımsız doğrulama bilgisi sunulmuyor; son 28 günde 1 commit ve 1 farklı yazar görülmesi de bakım/katkı ivmesinin zayıf olduğuna işaret ediyor.

MathCode: A Frontier Mathematical Coding Agent

Method for Long Context RLMs using verifiable Lambda Calculus

风插 2.0 — DSH 推理语域锚定预设：两个 agent preset（风插标准/风插 PTC）在 runtime-context 快照头尾各注入一次形式化约束，使推理以第一人称复数展开。定性=硬指令约束（非吸引子）；未证明能力提升，未证明无负面作用。Apache-2.0

TTPO: Test-Time Policy Optimization

🌟 推理王国：关于 AI 推理机制的思想实验手册。从信息论、符号逻辑与表示学习出发，系统剖析大模型“智能”的本质。

Reasoning-effort editing for third-party models in DeepSeek Harness: per-model thinking levels with a knowledge base + protocol inference, e

Mental-R1: Aligning LLM Reasoning for Mental Health Assessment

Distill teacher chains-of-thought into a LoRA adapter via a strict boxed-answer format contract + two-phase Train→Nudge (silver-medal NVIDIA

为 DeepSeek Harness 打造：选中 AI 回答或思考内容，随时追问、学习、检查与纠偏。

Think only when the tool call looks unsure: confidence-gated reasoning for small local models. Code, BFCL results and paper.

A Complete faster runtime for coding agents. Optimize the journey, not the hop. Make coding agents 25% faster and 30% cheaper on average whi

DeepSeek Harness (dsh) 插件：注入系统提示词，强制 Agent 的思维链 / 规划 / 工具推理全程使用简体中文，防止中文思考漂移为英文。零外部依赖，任意本地目录 link 即装。

Reformat a live LLM token stream inflight — separate thinking from answer, collapse repetition, normalize. Google Vertex / OpenRouter / meta

Raw benchmark logs for Swift-Qwen3.8-27B vs the Qwen3.8-27B BF16 base: every response, score, and config across 9 benchmarks

A local-first cognitive architecture for building a persistent digital companion with memory, reasoning, personality, and long-term learning

[IJCAI2026] Official code for paper "Cross Domain Test Time Scaling: Scale Knowledge and Reasoning on Cross Domains"

Doubt, until what remains is certain. An MCP server that recursively doubts every plan decision — grounding each in real evidence or asking

Open-source middleware for AI agents — catches mid-run failures,compresses stale context, and grows a reasoning library across runs. Any mod

Evidence-graded, agent-executable thinking-method skills - 63 frameworks reduced to their working mechanism, honestly graded, and producing

Persistent organisational intelligence for AI agents. Turn codebases, documents, policies and decisions into a knowledge graph — 13 LLM back

Bilingual graduate textbook on LLM post-training: SFT, RLHF, DPO, GRPO, RLVR, reasoning, agents, systems, evaluation, and safety.

Use GLM models in GitHub Copilot Chat with your own API key. Supports image input, tool calling, and streamed thinking.

Restores reasoning content dropped from multi-turn OpenCode conversations (DeepSeek, Kimi, GLM, MiMo, MiniMax, InclusionAI/Ling, OpenRouter)

AI reasoning skills distilled from 4,665 real Claude Fable 5 chain-of-thought traces. Mathematically tuned. Grade A (100%) emulation accurac

Open-source thinking-framework skills that make rigorous reasoning executable for AI agents — first-principles, inversion, second-order thin

Ground LLM output as provenance-tracked claims, not truths. A maximal super-ontology + native RDF-1.2 logic: reasoning engine — OWL, SHACL,