This project provides an autonomous agent for executing multi-step coding tasks, designed for developers needing to automate complex or repetitive development workflows. It enables continuous, iterative code generation and integration by orchestrating the Claude Code AI model, managing the entire pull request lifecycle, and maintaining persistent context across runs, thereby facilitating large-scale project completion and maintenance.

How It Works

The core mechanism involves running Claude Code within a continuous loop, inspired by CI/CD practices. Each iteration automatically creates a new Git branch, generates code commits based on a prompt, pushes the changes, and creates a GitHub pull request. The system monitors CI checks and reviews, merging successful PRs or discarding failed ones. Crucially, a shared markdown file (SHARED_TASK_NOTES.md) acts as external memory, preserving context and learnings between iterations, enabling the AI to self-correct and build upon previous work without losing state.

Quick Start & Requirements

• Installation: Execute curl -fsSL https://raw.githubusercontent.com/AnandChowdhary/continuous-claude/main/install.sh | bash.
• Prerequisites:
  • Claude Code CLI (authenticated via claude auth).
  • GitHub CLI (gh auth login).
  • jq (installable via brew install jq or apt-get install jq).
• Usage: Run via continuous-claude --prompt "your task" --max-runs <N> --owner <owner> --repo <repo> or --max-cost <USD>.
• Docs: Installation script, GitHub CLI authentication, jq installation instructions are provided.

Highlighted Details

• Fully automates the PR lifecycle: branch creation, commits, PR submission, CI check monitoring, and merging.
• Maintains context continuity across iterations using a notes file, enabling self-improvement and complex task decomposition.
• Capable of handling tasks like increasing test coverage, fixing dependency update breakages, and large-scale refactoring.
• Supports parallel execution of tasks using Git worktrees for concurrent development streams.
• Configurable execution limits via maximum runs (--max-runs) or a cost budget (--max-cost).

Maintenance & Community

The project acknowledges contributions and inspiration from individuals like Namanyay and concepts explored by GitHub Next's Continuous AI project. Specific community channels (e.g., Discord, Slack) are not detailed in the README.

Licensing & Compatibility

The project is released under the MIT License, permitting broad use, modification, and distribution, including for commercial purposes.

Limitations & Caveats

Failed iterations, while potentially "wasteful," are part of the learning process; the system discards work but learns from failures. The tool's effectiveness is contingent on the Claude Code CLI's capabilities and the clarity of the provided prompts. It requires prior authentication setup for both Claude Code and GitHub CLIs.