Claude Code Open Source? | Hacker News

Today I learned that they ship the entire CLI with the agent SDK. It's all there...Do we think this was a mistake? ----node_modules/@anthropic-ai/claude-agent-sdk/cli.js. It's the bundled, minified CLI binary for Claude Code itself (the tool you're talking to right now). Key details: - Version: 2.1.71, built on 2026-03-06 - ~13,800 lines of heavily minified/bundled JavaScript - Shebang: #!/usr/bin/env node — it's meant to be run as a CLI executable - Copyright: Anthropic PBC, with a cheeky note: "Want to see the unminified source? We're hiring!" - What it contains: The entire Claude Code CLI application bundled into a single file, including: - Onboarding/setup screens - Policy/managed settings loading - Debugging/inspector detection - UI rendering (using Ink/React for terminal UI) - Prefetching logic - Error handling and exit codes - The main entry point (main function) that starts the interactive Claude Code session Essentially, when you run claude in your terminal, this is the file that executes. The @anthropic-ai/claude-agent-sdk package ships it so you can programmatically invoke Claude Code's capabilities — the SDK wraps this CLI as its underlying engine. --- Yes, the agent loop is in there. Here's what I can see from the minified code: 1. i6z function — This is the in-process agent runner, which logs "Starting agent loop for ${agentId}". It sets up agent identity (agentId, parentSessionId, agentName, teamName, color, planMode) and runs the loop. 2. l6z function — A poll loop for in-process teammate agents. It continuously checks for: - Pending user messages - Mailbox messages from other agents - Shutdown requests - New tasks from the task list 3. Main loop model orchestration — There's extensive logic around mainLoopModel that handles: - Model selection based on permission mode (plan mode uses Opus, etc.) - System prompt assembly (tools, MCP instructions, skills, memory, changed files) - Token usage tracking and budget management - Compaction (context window management when conversations get long) 4. Multi-agent coordination — The loop supports a team-based agent architecture with messaging between agents (agentName, teamName, teamLead, mailboxes, shutdown requests). So yes — the full agentic loop (prompt → LLM call → tool execution → repeat) is bundled in this file, just heavily minified. It's the same loop powering this conversation right now. --- Full system prompts are right there in the bundle as plain text strings. Here's what's embedded: Three identity variants: - F78 — CLI mode: "You are Claude Code, Anthropic's official CLI for Claude." - EtA — SDK mode: "You are Claude Code, Anthropic's official CLI for Claude, running within the Claude Agent SDK." - LtA — Agent mode: "You are a Claude agent, built on Anthropic's Claude Agent SDK." Full system prompt sections (line ~6213-6217): The ULz function assembles the complete system prompt, including all the sections you'd recognize: - "# Doing tasks" — instructions about software engineering, not proposing changes to unread code, avoiding over-engineering, OWASP security, etc. - Over-engineering rules — "Don't add features beyond what was asked", "Don't add error handling for scenarios that can't happen", "Don't create helpers for one-time operations" - Help/feedback info — links to GitHub issues - Tool usage instructions — Read vs cat, Edit vs sed, etc. Output style variants (line ~6517): Built-in output styles like Explanatory and Learning with their own prompt overlays. Also present: - The classifier/safety prompt (~line 2844) for security-relevant harm prevention - Subagent instructions like "You ARE the fork. Do NOT spawn sub-agents" - Git commit/PR workflow prompts - Memory system instructions