wushenghuaandClaude Opus 4.7 988b8cc27b feat: show original task prompt in step detail view
Displays the task description (the original instruction sent to
Claude Code) in a highlighted box above the step timeline, so
users can always see what prompt produced the current task.

Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>
2026-05-29 16:11:09 +08:00

OpsPilot — AI Coding Agent Control Center

A task orchestration center for AI coding agents. Assign development tasks, let AI agents break them into steps, execute, and report progress — all visualized in a web dashboard.

Why OpsPilot?

If you use Claude Code, Cursor, Codex, or other AI coding tools across multiple projects, you face these problems:

  • Manually opening tmux sessions and launching AI agents for each project
  • Switching terminals to approve confirmation prompts mid-execution
  • No visibility into which project is doing what
  • Opaque task progress — you don't know what step the agent is on

OpsPilot solves all of this with a single web console for managing multiple AI coding agents simultaneously.

Quick Start

# 1. Clone
git clone https://github.com/wu736139669/OpsPilot.git
cd OpsPilot

# 2. Install dependencies (only FastAPI needed)
pip3 install fastapi uvicorn websockets

# 3. Start
python3 server.py

# 4. Open browser
open http://localhost:5016

Prerequisites: Python 3.9+, tmux, Claude Code CLI (claude command available)

Usage

Open http://localhost:5016 and:

  1. Click + New to create a project
  2. Use the directory browser to pick a project path
  3. Enter a task description
  4. The system auto-creates a tmux session, launches Claude Code, and tracks steps
  5. Click confirm buttons directly in the browser when prompts appear

Option 2: API

# Create a project
curl -X POST http://localhost:5016/api/projects \
  -H 'Content-Type: application/json' \
  -d '{"name":"MyProject","path":"/path/to/project"}'

# Assign a task
curl -X POST http://localhost:5016/api/tasks \
  -H 'Content-Type: application/json' \
  -d '{"project":"MyProject","title":"Build a REST API","steps":["DB schema","API routes","Tests","Deploy"]}'

Option 3: Let Another AI Agent Manage It

If you're in a different project with Claude Code, you can have it manage OpsPilot via API — create projects, spawn agents, and track progress programmatically.

Architecture

┌─────────────────────────────────────────────────┐
│                 OpsPilot :5016                    │
│  ┌──────────┬──────────────┬─────────────────┐  │
│  │ Projects │  Task Steps  │  Live Terminal  │  │
│  │ List     │  + Progress  │  + Controls     │  │
│  └──────────┴──────────────┴─────────────────┘  │
│          REST API + WebSocket                     │
└─────────────────────┬───────────────────────────┘
                      │ tmux send-keys / capture-pane
          ┌───────────┼───────────┐
          ▼           ▼           ▼
     ┌─────────┐ ┌─────────┐ ┌─────────┐
     │ tmux    │ │ tmux    │ │ tmux    │
     │ Session │ │ Session │ │ Session │
     │ proj-1  │ │ proj-2  │ │ proj-3  │
     │ Claude  │ │ Claude  │ │ Claude  │
     │ Code    │ │ Code    │ │ Code    │
     └─────────┘ └─────────┘ └─────────┘

Features

  • Multi-project management — Run multiple Claude Code agents in parallel, each in its own tmux session
  • Task step tracking — Auto-detect completion marks and update progress in real time
  • Remote confirmation — Detect Claude Code permission prompts and let you approve from the browser
  • Live terminal view — See agent output in real time with auto-refresh
  • Session management — Resume, kill, or create new Claude Code sessions per project
  • Directory picker — Browse and select project directories visually
  • Claude Code session history — List and resume previous Claude Code sessions

API Reference

See CLAUDE.md for the full API table.

Configuration

Variable Default Description
OPSPILOT_PORT 5016 Web server port
OPSPILOT_PROJECTS_DIR ~/Documents/Projects Default projects directory

License

MIT

S
Description
GitHub 镜像: wu736139669/OpsPilot(每日自动同步)
Readme
65 KiB
Languages
HTML 58.7%
Python 40.2%
Shell 1.1%