Dev builds 1,200-line Mini Harness that beats Claude Code on Terminal-Bench
PawelHuryn · x · 2026-11-05
Pawel Huryn open-sourced Mini Harness, a teaching-oriented agent harness in 1,200 lines of Python with only 2 dependencies. Unlike other minimal harnesses, it includes goal mode, permissions + auto mode, Agent Skills, MCP servers, resumable sessions, compaction, and even Anthropic's server-side compaction. On Terminal-Bench 2.0 it scored 9.2/10 with Claude Haiku 5.5 — beating Claude Code (8.4), mini-swe-agent (8.8) and OpenCode (8.2) — at just $0.11 per run, and matched Claude Code at 8.4 on DeepSeek V4.1 Flash.
Related event: Developer Open-Sources 1,200-Line Teaching Agent Harness in Python(4 posts)→
More from coding & agent
- tw93 open-sources Kami: a design system that makes AI-agent documents actually readable — sujingshen · 2026-11-05
- Power user shares the one thing he won't let Claude do: write his tweets — PrajwalTomar_ · 2026-11-05
- Old iPad reborn as a smart home wall panel, thanks to Claude — PrajwalTomar_ · 2026-11-05
- Running Claude Code on DeepSeek for chores: one job costs $0.0043 — PrajwalTomar_ · 2026-11-05
- Agent audited a 300K-star skill pack in 70 seconds, found startup hook and usage counter — PrajwalTomar_ · 2026-11-05
- Voice-first agent feedback: 20-minute screen-share review produced a full action plan — PrajwalTomar_ · 2026-11-05