Awesome AI Agents › Code Automation and Testing

AnandChowdhary/continuous-claude

⭐ 1380 Shell repository created 2025-11-15

Continuous Claude is a command line tool built around a Bash script that invokes Claude Code repeatedly instead of once, so that large multi-step projects can progress without a human driving each step. The author wrote it to raise unit test coverage on a very large codebase from zero to eighty percent, a task that one-shot agent runs handle poorly because the agent stops when it believes the job is done and has no opportunity for self-criticism or further improvement. The script acts as a conductor around the agent. In each iteration it creates a new branch, runs Claude Code to produce a commit, pushes the changes and opens a pull request with the GitHub CLI, monitors continuous integration checks and reviews through gh pr checks, then merges on success or closes the pull request and discards the work on failure, after which it pulls the updated main branch, cleans up and repeats. Because it rides on existing GitHub workflows, code owner approval rules, required checks and preview environments apply without extra setup. Continuity between iterations comes from a shared markdown file that serves as external memory: the agent records what it did and what should come next, and the prompt instructs it to treat the run as a relay race, making meaningful progress on one thing and leaving a clean handoff rather than verbose logs. The author reports that a vague instruction such as increase coverage evolves into a concrete procedure as the loop teaches itself through iteration. Suggested uses include dependency updates with follow-up fixes from release notes, and long refactors such as splitting a monolith into modules or modernising callbacks to async and await across many pull requests.

https://github.com/AnandChowdhary/continuous-claude

claude-codeautomationci-cdcoding-agentsgithub