Show HN: Watch-Claude-think – a CLI for watching Claude Code's thoughts (github.com)

Watch-Claude-think is an interactive CLI that gives a real-time view of Claude Code’s private “thinking” stream by tailing the local ~/.claude folder. You enable Claude Code’s thinking mode (Tab) and run it with npx (or pnpm/yarn/bun equivalents) or install it globally; it requires Node.js 18+ and reads active sessions from ~/.claude/projects. The tool makes no web requests and uses no tokens — it simply monitors and displays the model’s internal updates, letting you pick a project and either watch all sessions or focus on one (session names may not always match resume labels). For AI/ML practitioners this is a neat transparency and productivity tool: you can spot bad assumptions, excessive “spinning” (e.g., lots of “Actually…” rollbacks), over-mocking, or risky config edits before Claude applies them, and glean alternative approaches or dependency hints the assistant didn’t surface. It’s useful for prompt engineering, debugging multi-agent/code workflows, and faster iteration in code-heavy sessions. Because it reveals internal thoughts that don’t appear in final outputs, it also raises privacy/operational considerations about what context is being stored locally. The project is open-source (MIT) and aimed at developers who want to observe and intervene in Claude Code’s reasoning in real time.