Meet loopy, a terminal meta-agent that dropped on Hacker News this week with a deceptively simple pitch: stop prompting your AI coding assistants manually and start designing the loops that prompt them for you. Built by secretbuilds and currently targeting macOS users with Claude Code or Codex in their workflow, loopy runs as a background daemon using launchd, watching your session transcripts to surface patterns you'd never catch on your own.

The Manual Prompting Problem

Boris Cherny, creator of Claude Code at Anthropic, put it bluntly: "I don't prompt Claude anymore. I have loops running that prompt Claude and figure out what to do. My job is to write loops." That's the ethos behind loopy โ€” most developers are leaving 40-70% of their potential on the table by staying in prompt-response mode instead of graduating to autonomous automation. The tool claims a single well-chosen loop replaces 50-200 manual prompts per month, and three active loops can save you roughly 15-30 minutes daily.

How loopy Works

The architecture is straightforward: a watcher daemon monitors your Claude Code sessions via launchd, passing transcripts through a digester that compresses and redacts them before sending compact summaries to your own local claude -p binary. The engine analyzes these digests for recurring patterns, then drops promising candidates into an inbox as proposals โ€” you review, approve or dismiss, and approved loops become self-contained bundles with their own trigger logic and operating instructions. Everything stays on your machine; loopy never phones home. It even redacts API keys, tokens, passwords, and high-entropy strings that look like secrets before any LLM call happens.

The /fable Integration

One interesting detail buried in the docs: the installer also drops a /fable slash command into ~/.claude/commands/, routing your prompt through Claude Fable 5's system prompt inline within your current session. It spawns claude -p --model claude-fable-5 as a subprocess and returns the output without leaving your active context โ€” clever for those moments when you need Fable's specific personality without switching models entirely.

Privacy by Design

Transcripts never leave your machine. Before any digest reaches an LLM, loopy scrubs credentials, GitHub tokens, AWS keys, URL credentials, and anything else that looks sensitive. The engine only sends redacted summaries to your own claude -p process โ€” no external services, no subscriptions, no cloud component beyond your existing API credits.

Getting Started

Installation is a single curl command: curl -fsSL https://raw.githubusercontent.com/secretbuilds/loopy/main/install.sh | bash. Requirements are minimal โ€” Node โ‰ฅ 20, git, the claude CLI in your PATH, and macOS. The setup process links the binary to your PATH and configures the background daemon; optional flags include --companion manual for no automatic nudges or --no-daemon if you want to configure without the watcher running. Once installed, running loopy opens a full-terminal dashboard with three panels: inbox (pending proposals), loops (installed automations), and activity (scrolling log of everything the daemon has done).

Key Takeaways

  • Loopy is early-stage software โ€” Windows/Linux daemon support is on the roadmap but not here yet
  • Everything runs locally using your own claude -p binary; no separate service or subscription required
  • The dashboard needs at least 60ร—16 terminal dimensions to render properly
  • Early users report discovering 3-5 automatable patterns per month they wouldn't have spotted manually
  • Boris Cherny's reported output with loops: 8-15 PRs shipped daily without touching code himself

The Bottom Line

Loopy is exactly the kind of tooling that signals where AI-assisted development is heading โ€” not more prompting, but meta-systems that handle the repetitive stuff so you can focus on architecture and judgment calls. If you're still manually typing prompts into Claude Code in 2026, you're working harder than you need to. The writing's been on the wall since Cherny's workflow post went viral; loopy just makes it accessible to anyone willing to hand over a bit of control for a lot less cognitive overhead.