How to get an AI coding agent to open pull requests from GitHub issues

Put the issue in GitHub, add a trigger label, and let a runner implement it in an isolated workspace and open a pull request linked back to the issue. ProPR does exactly that with Claude Code, Codex, Antigravity CLI, Mistral Vibe or OpenCode, and your team reviews the result like any other pull request.

What an AI coding agent that opens pull requests automatically needs

Any agent can edit files. Turning a GitHub issue into a reviewable pull request takes four more things, and each one is a design decision whichever tool you pick.

A trigger the team already uses. An issue label or a comment starts the work, so nobody learns a new inbox.
One workspace per task. The agent gets its own checkout, branch and container, so parallel tasks never collide and your main checkout never changes.
A pull request as the only exit. The runner commits, pushes and opens the PR, and the agent holds no write access to GitHub.
A record attached to the change. The prompt, the log, the model and the cost sit next to the PR, so a reviewer sees why the diff looks the way it does.

ProPR builds in all four.

Label the issue

Install ProPR on a Linux Docker host and add the repository. Two commands bring the stack up; setup signs in your agents and connects GitHub in the browser. The quickstart walks through it.

npm install -g propr-cli
propr setup
propr repo add owner/repo

Write a normal GitHub issue and add the AI label. A model label such as llm-claude-opus55 picks the agent and model, and several model labels on one issue produce one pull request per model, so you compare their answers side by side. ProPR picks the issue up, tracks progress with a state label, and streams the run on the dashboard: each edit explained, each test run.

The execution trace of a live run in plain language: why the time-zone fix is needed, the new DST test cases, and all 14 checkout tests passing — The live log of a running task: what the agent changed and why, and the tests it ran.

The run opens the pull request

The agent writes code in its own Docker container, Git worktree and branch, with a read-only GitHub token. ProPR validates the result, commits, pushes and opens a pull request linked to the issue. How isolated runs work.

The pull request carries the diff and a run summary: status, duration, token cost and model. The task view keeps the full prompts, the live log and the per-call cost, in the web UI and the CLI.

Task header: repository and pull request, Implementing status, the run in progress, its model and runtime, and token consumption — The run header on the task: status, duration, token cost and the model that ran.

Review, comment, merge

Review the diff in GitHub the way you review any pull request. Write what you want changed as a plain comment, in the conversation or on a diff line, and ProPR pushes one more commit to the same branch and replies with links to the task, the commit and an undo. How PR refinement works.

/review posts numbered merge blockers, suggestions and a score without touching the code, and /fix applies the findings you keep. You merge in GitHub. Every comment command.

Bigger changes start with a plan. Describe the change, check the context ProPR gathered, approve the task breakdown, and the issues come out of the plan ready to run.

A follow-up run on the pull request: read the PR comment about the phone layout, move the summary below the form, run the checkout tests — A follow-up run started from a pull request comment.