Dedicated Git worktree
Each task gets its own checkout. Parallel tasks never collide, and your main checkout never changes.
Every task gets its own Docker workspace, Git worktree and branch. The agent edits there. ProPR validates, commits, pushes and opens the pull request.
Each task gets its own checkout. Parallel tasks never collide, and your main checkout never changes.
The agent runs in a standard container with a read-only GitHub token. Optional restricted network mode limits it to an allowlist of hosts.
Each task gets its own branch. ProPR owns every commit, push and cleanup.
The change reaches your team as a pull request, with your checks, approvals and merge policy in GitHub. The guide to getting an AI coding agent to open pull requests from GitHub issues follows one task from label to merge.
New Task sends one instruction straight to implementation. Describe the change, attach files, and run it. ProPR files the GitHub issue, runs the task in isolation and links the pull request back to it.
Broad or risky changes start in planning. Longer objectives run as Goals.
For each approved plan, pick one Epic PR or a pull request per issue, the agent and model, Ultrafix and auto-merge.
The exact prompt, agent logs, model usage and file changes stay on the task after the run ends. Reviewers see what the agent did, in the web UI or the CLI, and what each AI-written pull request cost.

The task page streams the run live: why the agent makes a change, the files it edits and the tests it runs. The trace stays on the task afterwards.
Visual previews put a screenshot or video of the changed screen on the pull request, and every follow-up commit posts fresh ones.
Choose images, videos or both, and tell the agent which pages or viewports to show.
The agent records the running change with the tooling your repository already uses, such as Playwright or Storybook.
The web UI and desktop app show private PR media, and Connect Plus keeps the full-resolution originals.
Commit a .propr/ folder and every run follows it. Change a rule in a pull request and the next run picks it up. propr init scaffolds the files.
.propr/workflow.ymlSetup hooks, agent instructions, the validation commands ProPR runs after the agent exits, a parallel task limit, a per-run spend cap and the network mode.
.propr/workflow.yml and the hosts it denied..propr/pr-template.mdShape the title and body of every pull request ProPR opens, with placeholders for the issue, model, cost and changed files.
The auto_merge block picks the merge method and lists the paths that always wait for a person.
Once the PR opens, /review posts structured findings and a score before a person looks.