Think in Code — MANDATORY Paradigm
LLMs are now instructed to program analysis instead of computing it mentally. When data needs to be analyzed, counted, filtered, compared, or processed — the LLM writes code via ctx_execute and console.log() only the answer. Raw data never enters context.
One script replaces ten tool calls. 30 lines of code saves 500KB of context. Your limit lasts 10x longer.
What changed
- 3 MCP tool descriptions updated:
ctx_execute,ctx_execute_file,ctx_batch_execute— all now include Think in Code instructions - 4 routing block messages updated: curl/wget, inline HTTP, build tools, WebFetch — all redirect with Think in Code guidance
- 12 platform instruction files updated with platform-correct tool prefixes: Claude Code, Cursor, Codex, VS Code Copilot, Gemini CLI, Kiro, Zed, OpenClaw, OpenCode, KiloCode, Pi, Antigravity
- Root CLAUDE.md updated with Think in Code manifesto
Rules enforced
- Pure JavaScript only — no npm dependencies
- Node.js built-ins only (
fs,path,child_process) - Always
try/catch, null-safe - Node.js + Bun compatible
console.log()only the answer
Example
WITHOUT Think in Code:
Read(file1) + Read(file2) + Read(file3) = 150KB in context
LLM processes 150KB mentally = token burn
WITH Think in Code:
ctx_execute("javascript", `
const fs = require('fs');
['file1','file2','file3'].forEach(f => {
const lines = fs.readFileSync(f,'utf8').split('\n').length;
console.log(f + ': ' + lines + ' lines');
});
`)
= 100 bytes in context. Same result. 1500x less tokens.
Full Changelog: v1.0.63...v1.0.64