harness2

Harness² CLI: a general-purpose version of Harness² (google-research/harness2). Tune any coding agent's harness (CLAUDE.md/AGENTS.md, skills, subagents, hooks) on your own tasks, with your Claude or ChatGPT subscription or any model provider.

为 DeepSeek Harness 注入递归自化能力优化，以及可视化调整

Minecraft Agent for DeepSeek Harness Web/Desktop: live game view, native reasoning and tools, configurable personas, server notebooks and local survival knowledge. / 我的世界智能体

Harness² on Pi Durable: runs a task, tests harness changes on fresh copies, adopts the best, and learns lessons across tasks — model weights stay fixed.

Harness² experiments for DeepSeek Harness: isolated adapters, measured comparisons, native experiment views and evidence-linked lessons.

Evidence-first document review with real LangGraph workflows, a custom Harness² adapter, durable human approval, and a transparent interactive walkthrough.

FsocietyHarnessHack — fork WIP de Harness² para seguridad ofensiva autorizada y simulación. No afiliado a Google ni a ningún Estado.

Convert a Google Research Harness2 harness tree into a DeepSeek Harness (DSH) agent-preset bundle

Streaming LangGraph coding agent with OpenRouter, resumable terminal chat, and Harness² refinement

Anti-hallucination engineering harness. risk = confidence^2 x (1 - evidence). i^2 = -1, implemented. MIT.

Data pipelines that improve without retraining: Google Research's Harness² loop applied to five data-engineering tasks, with a tested answer wall and a blind-spot experiment

Harness² as a Pi extension: recursively refine the harness Pi runs under for the task in front of you

Read pi-harness2 lessons from Pi and DeepSeek Harness

A Rust-native implementation of Google's Harness², featuring recursive agent harness optimization, adaptive execution, and cross-task learning. Powering AstraAeon.

Harness² recursion: two auditable dsh plugins — task-level probe/compose evidence loop + harnessing-level experience library (frozen contract v4, 34/34 tests, real-session E2E)