#reasoning

Qt/C++ OpenAI-compatible reasoning degradation guard proxy with GUI, CLI, deb and AppImage packaging

Keyless heuristic self-inspection tool: an agent describes a thought or task and always gets back one metathought. Deterministic routing over an open CSV (no LLM). REST + MCP.

Open-source Agent Skill for Claude Code that makes Claude reason with explicit probabilities, base rates, and falsifiers. Distilled from 51 books; first in 18/18 blind passes vs baseline Claude and GPT-5.5, and benchmarked against 14 rival skills and prompt techniques with full cost telemetry (round 5).

Judge-gated AI task orchestration: context recovery, real subagents, preflight reviews, repair loops and verified delivery.

Code and data for "When Does Reasoning Age? Survival Analysis of Step-Level Error Hazard in LLM Chains" (EMNLP 2026)

A rigorous four-agent research workflow for Codex with evidence mapping, concrete advancement, independent review, and accountable synthesis.

Claude Newton: AI Reasoning Partner 2026 - Calibrated Pushback & Current Sources

A Claude skill that makes stuck arguments, drafts, plans and decisions click into place: visualize every piece, rotate the shape until it locks, then rebuild.

让 DeepSeek 变成一只疯狂偷吃用户 token 的大肥鱼 🐟 —— 最大思考模式（Max Thinking）for DeepSeek Harness：面对没有现成答案的问题，多候选推演 × 证据优先 × 六视角审视 × 隔离对抗审思，锻造出被验证过的突破方案。代价：token 吃得又快又多。

Privacy-first local Ollama chat with expert personas, reasoning, multimodal input, context-aware history, and long-term conversation memory.

Use CrofAI models in GitHub Copilot Chat with your API key.

Fully offline, portable AI agent for Windows/Linux/macOS - llama.cpp launcher with Web UI, OpenAI-compatible API, agent tools, MCP (web search) and reasoning. One-command setup. No cloud, no API keys.

Reproducible QLoRA SFT and DPO experiments for open-weight LLM reasoning, with locked selection, matched controls, and audit-grade provenance.

LocalSight · 198M-A64M 思考型 MoE LLM，2×RTX4090 从零训练（pretrain→SFT→SimPO→RLAIF→Agent RL），GGUF/Ollama 可运行

Agent skills for honest reasoning (dialectic audits, witness independence, loop closing, a pre-registered campaign gate) and practical tools. Apache-2.0.

World-model agents for ARC-AGI-3: online rule induction, uncertainty probing, and planning — research prototypes