Google open-sources Harness², letting agents recursively optimize their own harness without touching model weights
aigclink · x · 2026-10-18
Google has open-sourced Harness², a plug-and-play framework for recursive agent harnessing that lets agents automatically improve their own execution system — prompts, tools, memory, skills, subagents, scripts, plugins, orchestration — without training the model or touching weights.
It pairs two levels of recursion:
- Task-level recursion: a contrastive Propose–Probe–Compose loop — propose harness edits, probe them by running the same task under different configs, then compose a refined harness from behavioral contrasts between executions.
- Harness-level recursion: after each task, the adopted harness and evidence-backed improvement lessons are retained to guide future optimization.
It supports parallel and sequential recursion, works with executors like OpenCode and Codex, and optimizes eight component types. The repo has racked up 3k stars on GitHub.
More from coding & agent
- Hermes Agent + local models manage all machines; hermes-skins repo hits 607 stars — Teknium · 2026-10-19
- Harness is dead? AI Engineer NYC debates stripping scaffolds vs. multi-day agents — TheTuringPost · 2026-10-19
- Open-source Claude Code plugin routes grunt work to cheap models, saves Opus for judgment — investigatormaker · 2026-10-19
- Developer builds a standalone Mac app from scratch in 1.5 hours with Cursor — gaganghotra_ · 2026-10-19
- Apify founder Jan Curn: every agent payment protocol exists, real volume is near zero — AI Engineer · 2026-10-19
- How I imported 10,000+ media notes into Obsidian with AI, safely — dSebastien · 2026-10-19