mini-harness: a ~1,200-line teaching agent harness with MCP, skills and compaction
PawelHuryn · x · 2026-11-05
Pawel Huryn open-sourced mini-harness, a teaching-oriented agent harness in 1,200 lines of Python with only two dependencies—small enough to read in one sitting.
It covers features most minimal harnesses skip: goal mode, permissions, Agent Skills, MCP servers, resumable sessions, subagents, and compaction including Anthropic's server-side compaction. It runs on Claude and any model speaking Anthropic's message format, is MIT-licensed, and comes with a newsletter that walks readers through building the full harness step by step with copy-paste prompts.
Related event: Developer Open-Sources 1,200-Line Teaching Agent Harness in Python(4 posts)→
More from coding & agent
- Aether launches cloud runtime for Claude Code and Codex: 19.4s startup, no second AI bill — rohanpaul_ai · 2026-11-05
- Cohere Labs talk: Kernel Autoresearch uses coding agents to auto-discover valid ML kernels — Cohere_Labs · 2026-11-05
- Don't pride yourself on minimal tokens — your time is a cost too, developer argues — brandon_galang · 2026-11-05
- After Agent-Tweaked His Site for AI Search, ChatGPT Drove 57% of Traffic — jamesbrooksco · 2026-11-05
- Developer says Codex no longer worth paying for, wants a model better than Opus 5.5 — weswinder · 2026-11-05
- 83% of orgs run unapproved AI coding agents; 56% have agents acting under own identity — emmanuelvivier · 2026-11-05