Insights
Gated Dependencies: Making LLMs Do What You Say
When an LLM ignores your 'ALWAYS do X' instruction, don't add more emphasis. Restructure so the downstream behavior requires X's output. Make the tool call load-bearing, not decorative.
6 Harness Engineering Patterns I Found in Claude Code's Source
I studied the leaked Claude Code CLI source, 512K lines of TypeScript that wrap Claude into an autonomous coding agent. Here are the patterns worth stealing for your own AI agent harnesses.