Context reducer for DeepSeek Harness! I always wanted to intervene in how Claude Code handles context. Early days you could edit the session JSONL, that door closed. So I read how @deepseek_ai Harness cuts big tool output, then Codex, then pi. All three: keep head, keep tail,
Author of Crawl4AI. AI & software builder, systems thinker, founder of Kidocode. I build, analyze, and advise on hard technical problems.
- Claude Code forgets my rules past 100k tokens. CLAUDE.md is read once and never again, so the ASD-STE100 rules I added yesterday went too. So I built a hook. It re-injects my rules every 64k tokens, not every prompt. Zero tokens between steps. Install file in the reply. Hand it
- Since Opus 4.6 its English became very hard to understand. Coding abbreviations, metaphors, made-up terms. I added ASD-STE100 Simplified Technical English to my CLAUDE.md, thanks @bil0090. His prompt is in the reply. Finally something useful comes out. Funny to pay $200 then
- Rewriting Crawl4AI in Rust. To actually learn the language (not just read about it), I built Rustchap - 477 puzzles, every answer precompiled with real rustc so it runs offline and the compiler is the judge. Deliberate practice > only books, at least for me. Anyone else made
- DeepSeek just dropped Harness: a fully modular, plugin-based AI framework. Swap models, tools, or loops without touching core code, perfect for rapid experimentation. MIT, and that makes it to grow in community way. deepset.ai/harness🧩 DeepSeek Harness v0.1 is now available in Developer Preview! 🔹 We’re opening it up to developers building agent harnesses worldwide and open-sourcing the codebase in MIT license. 🔹 Powered by the Cordis meta-framework, DeepSeek Harness is an agent harness built around one


