Continuous Claude

Continuous Claude is an open source command line tool that runs Claude Code in a continuous loop, so an AI agent can keep working on a large task across many iterations instead of stopping after one attempt. Its author, Anand Chowdhary, built it to raise test coverage on a codebase with hundreds of thousands of lines, and it suits any long running job that can be split into small, reviewable changes. Each iteration follows a pull request workflow. The script creates a branch, runs Claude Code to produce a commit, pushes it and opens a pull request with the GitHub CLI, watches CI checks and reviews, then merges on success or closes and discards the work on failure before pulling the updated main branch and starting again. Because it rides on existing GitHub workflows, required reviews, code owner approvals and CI checks are all respected. A shared markdown file acts as external memory where the agent records what it has done and what should come next, which carries context between otherwise stateless runs. Runs can be capped by iteration count, estimated cost or duration, and the tool can also drive OpenAI Codex. The software is free and released under the MIT license. Users pay only for the underlying model usage through their own Claude or Codex access, and the tool reports the cost of each iteration as it runs. Failed iterations are thrown away by design, which spends tokens but lets the next attempt learn from the recorded test failures.