Claude Code · Codex · PR Operations · Mac Studio · Parsec

25+ terminals. Continuously running agent loops. 300+ PR merges a day. Around 30,000 commits a month.

We're hiring someone to operate this workflow for Vedika at XALEN—and keep the software working while the agents keep working.

Internally, call it the Pro Max Operator.

The job is part AI-agent supervision, part software engineering, part integration, and part incident response. PR babysitter describes approximately 5% of it.

You are not here to watch terminal windows or click Merge all day. You are here to know what every agent is doing, catch bad changes, recover stuck work, and keep the operation moving without losing control of quality, cost, or scope.

The environment

Multiple Claude Code and Codex sessions running across 25+ terminals, persistent and nested execution loops, parallel branches, overlapping changes, and a high-volume PR queue.

Your workstation is a Mac Studio, with remote access through Parsec.

Some agents will finish correctly. Some will get stuck. Some will pass their tests while breaking somebody else's changes. Some will turn a small fix into an unsolicited architectural renovation.

You need to know when to let them continue—and when to stop the loop.

The volume describes the workload. It is not a quota for blindly approving changes.

What you will own

1. The agent fleet.

Launch, supervise, pause, resume, and recover parallel coding sessions. Know which task, branch, and workspace each session owns. Catch idle agents, duplicate work, lost context, and agents confidently solving the wrong problem.

2. The loops and orchestration.

Make long-running automation observable and controllable. Maintain checkpoints, retry limits, timeouts, resource limits, and clear stop conditions. Stop a failing loop before it produces another hundred failures.

3. The PR queue.

Triage incoming changes, inspect diffs, enforce checks, coordinate reviewers, resolve conflicts, sequence dependent PRs,



and merge eligible work without breaking the main branch.

4. Git and parallel integration.

Manage branches and worktrees, recover interrupted work, handle rebases, prevent agents from overwriting each other, and reconcile competing implementations. Understand the changes—not just how to remove conflict markers.

5. Scope and task quality.

Turn broad requests into bounded tasks with acceptance criteria. Stop a small fix from becoming an unrequested rewrite. Separate useful follow-up work from agent-generated scope creep.

6. Testing and build reliability.

Reproduce failures, distinguish product bugs from environment problems, investigate flaky tests, and verify that separately passing changes still work together. A green check is evidence, not permission to stop thinking.

7. Mac Studio and remote operations.

Keep processes, storage, memory, toolchains, and Parsec access usable. Handle runaway processes, disconnected sessions, full disks, broken environments, and restart recovery without unnecessarily destroying healthy work.

8. Security and spending.

Protect credentials, maintain appropriate permissions, scrutinize unexpected dependency and configuration changes, and track usage, quotas, and wasteful retries. Know which actions require approval before an agent executes them.

9. Releases and recovery.

Prepare releases, verify integrated behavior, coordinate higher-risk changes with the responsible engineer, and maintain an actionable rollback path. Own the operational response when a merge or release goes wrong.

10. Making the operation less dependent on babysitting.





Build scripts, dashboards, alerts, and runbooks. Automate recurring interventions. Keep handovers explicit enough that another operator can take over without reconstructing the entire situation from terminal history.

What you need

You should be able to read and debug real code—not just prompts.

Strong Git skills, CI troubleshooting, shell scripting, process management, log analysis, and failed-build recovery are essential. You should be comfortable operating Claude Code and Codex beyond a single interactive session.

Experience with parallel agents, worktrees, high-volume repositories, macOS development environments, and remote workstation operations is especially relevant.

You do not need to know every framework. You do need enough engineering judgment to recognize when a problem requires a specialist instead of another retry.

Show us systems you have operated, automation you have built, and failures you have recovered from.

What good performance looks like

A healthy main branch. Fewer regressions and duplicate PRs. Predictable merge queues. Clear ownership of active work. Faster recovery from failed sessions. Visible resource costs. Less manual intervention for the same amount of verified delivery.

More commits is not automatically better performance. More working software, with less operational chaos, is.

The operator must have authority to pause agents, block unsafe merges, and escalate architectural, security, or release decisions. Continuous system operation requires explicit coverage arrangements—not an assumption that one person is available 24/7.

Apply

Send your GitHub profile or relevant work samples, a short description of an automation workflow you have operated, and one concrete example of recovering a broken build, agent run, or integration.

- Tell us how you would get visibility and control over 25+ sessions before increasing concurrency further.

📌 PR Babysitter Pro Max AI Operator (Pune)
🏢 Vedika API
📍 Pune