# context-engineering

Every two hours, Codex drafts email replies for review. The drafts are good — he edits them anyway. Those edits are context too, and most automations throw them away. The fix: an inner loop brings context to the work; an outer loop recovers context from the review diff.

A context window is a model's day: how many lessons, messages, tool results, and task events Ryland can experience before sleep, compression, or collapse.

Cognition's Walden Yan returns after Don't Build Multi-Agents with three patterns that actually ship: Devin Review's clean-context loop, cross-frontier smart friends, and manager Devin. The common rule: writes stay single-threaded.

OpenAI DevX's Dominik Kundel says Chronicle means he no longer packages context for AI: one line can sync docs, edit markdown, open a PR, and DM Slack. Nice, but Chronicle's costs are real: screen recording, unencrypted local memories, and prompt-injection risk.

A Tsinghua Shenzhen team proposes Natural-Language Agent Harnesses: move agent control logic from code into structured language executed by an IHR runtime. Harnesses can reshape behavior, but more structure does not always mean better results.

Tired of tweaking prompts and swapping models while agents still fail to evolve? This post shows a simple Markdown context system that turned one person's agents from clumsy interns into autonomous powerhouses in 40 days, without changing models.

The core message: most people do not fail because the model is weak; they fail because context management is messy. Start with a minimal CLI workflow, then iterate through rules, skills, and clear endpoints until agent behavior converges.

A Sully.ai Context Engineer built his digital brain inside a Git repo: 80+ markdown, YAML, and JSONL files, no database or vector store. Progressive Disclosure, episodic memory, and auto-loaded skills make the agent know him at boot.

An AI Agent burns 34,500 tokens of system prompt every single conversation turn. The author used layered loading (always-on vs on-demand) plus a dual-model strategy to cut monthly costs from $568 down to $120-150 — a 75% reduction. Full breakdown with real numbers inside.

The 'Context Tax' in AI brings triple penalties: cost, latency, & reduced intelligence. Nicolas Bustamante's 13 Fintool techniques cut agent token bills by up to 90%. A real-money guide for optimizing AI context, covering KV cache, append-only context, & 200K token pricing.

Junde Wu built OneContext after getting fed up with coding agents forgetting between sessions. It uses filesystem, Git, and knowledge graphs to work across sessions, devices, Claude Code, and Codex; the GCC paper hits 48% on SWE-Bench-Lite.

For vibe note-taking to work well, you must force Claude Code to be 'picky.' Use a 4-layer filtering mechanism (file tree → YAML descriptions → outline → full content) to make it more selective. This pattern is called Progressive Disclosure.