I wrote a small open source project to orchestrate dockerized Claude Code sessions from Jira, GitLab, and Forgejo. It runs Claude in a container, with a branch per issue. Then automatically opens a PR, responds to CI status, and incorporates your PR feedback.

Before this, I was using Claude in Git worktrees, but I didn't like the fact that it was running on my machine unchecked. So I set strict permissions, which nerfed Claude’s autonomy.

For reviewing, I prefer the GitLab interface with a full PR, rather than the default accept/decline change-by-change flow while Claude is writing (maybe a personal preference).

Optionally, there is a knowledge base to keep context in. This is a workflow I've been using for a while: Create a repo with markdown files describing project best practices and preferences, then feed it to Claude as a simple search MCP. The orchestrator will automatically open PRs on that repo based on review comments on PRs once they're closed.

It's a work in progress, but seeing that OpenAI did something similar for Codex internally, I thought it might be useful for other Claude Code users as well. Happy to take PRs with improvements or ideas.