Continuous Claude

Detailed Introduction

Continuous Claude is a lightweight CLI that runs Claude Code in a loop to perform iterative development: each iteration generates changes on a branch, creates a pull request, waits for CI checks and reviews, and merges when checks pass. The project uses a shared markdown file (default SHARED_TASK_NOTES.md) as persistent context so subsequent runs remember previous decisions and outstanding tasks, enabling a relay-like autonomous workflow. It depends on the GitHub CLI (gh) and access to Claude Code, making it suitable for long-running, incremental improvements on large repositories.

Main Features

• Loop-driven execution with configurable max runs or cost budget.
• Full PR lifecycle automation: branch, commit, PR, monitor checks, merge or discard.
• Persistent context via a shared notes file to reduce context drift between runs.
• Parallel execution support via git worktrees for concurrent subtask processing.

Use Cases

Ideal for large-scale, long-term repository work such as incrementally adding unit tests, fixing post-upgrade breakages, large refactors, or automated style migrations. Teams that want to hand off repetitive engineering tasks to models while keeping human review via PR/CI will find Continuous Claude acts as an enhanced Dependabot or automation assistant.

Technical Features

• Shell/CLI-first design, easy to run in CI, containers, or developer machines.
• Integrates Claude Code with gh to wire model outputs into standard GitHub workflows.
• Supports --disable-commits dry-run mode for safe validation.
• MIT-licensed and documented on the project homepage for research and experimentation.