prompter logoprompter

One command. Expert AI agent. No prompt engineering.

Install
curl -fsSL https://prompter-7fe6d.web.app/get.sh | bash

Requires git, node, and python3. Works on macOS and Linux.

See it in action

~/ cd my-project ~/my-project prompter
prompter ──────────────────────────────────────── Workspace: ~/my-project Agent: codex
/help for commands, /settings to configure
❯ Fix the auth middleware returning 401 for valid tokens
──────────────────────────────────────────────────────────── Category: Bugfixer Expert: Auth & Security Expert Agent: codex ────────────────────────────────────────────────────────────
[E] Execute (execute directly) [P] Plan (research + plan, then execute) [L] Loop (split into tasks, execute one-by-one)

How it works

1

You describe the task

Paste a bug report, feature request, or plain English description. Supports images too.

2

Phase 1: Expert prompt generation

An AI agent reads the relevant source files, selects the right category, and writes a detailed execution prompt grounded in your actual code.

3

Phase 2: Execution

A specialist agent executes the prompt — making changes, running tests, and optionally committing and pushing.

Execution modes

E Execute

The default mode. Prompter generates an expert prompt from your request and hands it straight to the agent for immediate execution. Best for focused tasks like bug fixes, small features, or refactors where you know what you want.

Mode [E/p/l]: E
→ Agent reads code, makes changes, runs tests, done.

P Plan

The agent investigates your codebase first — reading files, tracing call chains, searching for patterns — and produces a detailed plan before making any changes. Once the plan is ready, it executes end-to-end. Ideal for complex tasks where you want the agent to think before it acts.

Mode [E/p/l]: P
→ Phase 1: research & plan. Phase 2: execute the plan.

L Loop

For large, multi-step work. Prompter breaks your request into discrete tasks, then executes them one by one — each in a fresh agent session with full context. Every completed task is committed and pushed automatically. The agent restarts between tasks to avoid context window limits and maintain focus.

Loop plan: 12 tasks (agent restarts between each task)
[1] Extract auth middleware into separate module
[2] Add input validation to API routes
[3] Write integration tests for auth flow
...

Built for real workflows

AI Multi-agent support

Prompter works with multiple AI agent backends. Switch between Codex, Claude Code, and Gemini CLI with a single setting. Each agent gets the same expert-crafted prompt, so you can pick the best tool for the job or compare results.

~ Auto-discovery

On first run in a new project, prompter scans your codebase and discovers expertise categories automatically — no configuration needed. It identifies your tech stack, directory structure, and testing patterns to route future tasks to the right specialist.

? Prompt review

Before execution starts, you can optionally inspect the generated expert prompt. Review what the agent was told, regenerate it if it missed the mark, or abort entirely. Full transparency into what your agent will actually do.

↗ Git integration

After completing a task, prompter can automatically commit and push changes. In loop mode, each task gets its own commit with a descriptive message. No manual git workflow needed — just review the results.

🖼 Image support

Drag and drop screenshots, mockups, or error captures directly into your prompt. Prompter extracts image paths and passes them to the agent, so it can visually understand what you're describing.

Prerequisites