AutoHarness: a self-learning skill layer for Claude Code hits 5.2k stars
adnan_hashmi · x · 2026-09-28
AutoHarness, an open-source project with 5.2k GitHub stars, adds a self-learning skill layer to Claude Code:
- Distills skills from your real sessions, merges same-scenario ones instead of stacking near-duplicates, updates them in use, and prunes unused ones — no daemon, no benchmark required.
- Same model, different harness: 42% → 78% on CORE-Bench (HAL), echoing swyx's "Big Model vs Big Harness" argument that harnesses do much of the work yet get rebuilt by hand every model generation.
- Distributed as a Claude plugin with hooks, agents, MCP config, and tests.
More from coding & agent
- Redditor post-trains an 80B MoE base into an agentic model on 4 V100s, livestreamed — jjusko20 · 2026-09-29
- Aegis: a security triage agent that remembers past analyst decisions — Certain_Fun_906 · 2026-09-29
- Manus Cue Agents Run as Independent Entities With Own Number, Inbox and Card — ChrisUniverse · 2026-09-29
- Dev: 'Showing your prompt' is now meaningless — context is references, skills and examples — trq212 · 2026-09-29
- Dev rebuilds childhood arcade game Off-Road with Codex and GPT-6, real spinning wheel controller — CSProfKGD · 2026-09-29
- AI voice agents now settle medical bills in one call — voice is the easy part — alex_verem · 2026-09-29