AI Agent Runner — Monitor Sessions

DESCRIPTION We are running a large parallel build using multiple Claude Code agents working simultaneously on a Laravel 13 / Livewire codebase. We need operators to keep those agents moving around the clock. This is not a development role. You will not architect, design, or write features. You are the human in the loop who keeps agents unblocked and escalates anything that requires a decision. WHAT YOU WILL ACTUALLY DO Monitor 6 to 8 Claude Code sessions through a Slack channel. No SSH, no terminal, no local setup — everything runs through Slack, so a laptop or even a phone works. Answer routine agent prompts: rerun the suite, retry a transient failure, restart a stalled session, clear a bloated context. Read PHPUnit and Pest output, PHPStan errors and lint failures, and identify which session they belong to. Escalate anything involving scope, architecture, contracts or approvals to the lead, with a clear one-paragraph summary. Keep a short written status log per session and hand off cleanly at shift change. Sweep your assigned sessions on a fixed cycle rather than reacting to whatever pinged last. THE SINGLE MOST IMPORTANT QUALITY Knowing when NOT to answer. Agents ask confident-sounding questions that are actually architecture decisions. The best person in this role recognises those instantly and escalates instead of improvising. We would far rather you escalate ten times too often than guess once. The codebase is protected by automated checks, and a wrong guess that passes them is more expensive than one that fails. REQUIRED Hands-on experience running Claude Code, Cursor, Codex, Aider or a similar agentic coding tool, on real work rather than tutorials Enough PHP and Laravel to read a stack trace and understand what broke and where Comfortable reading a git diff and understanding branches. You will not be pushing code. Can read test output and static-analysis errors and summarise them accurately Written English clear enough to escalate concisely in Slack Reliable for scheduled shifts. This is time-boxed work with a hard deadline. NICE TO HAVE Livewire experience Have run several agents in parallel before QA, DevOps or release-engineering background Horizon, Redis or queue debugging NOT REQUIRED, AND NOT WANTED Systems architecture or design experience. You will not be making those calls. A full-stack portfolio. This is an operations role. Any desire to improve the plan. The plan is fixed and enforced by automated checks. TERMS scheduled shifts. Tell us your timezone and which hours you can commit to. Hourly, tracked through Upwork NDA required before any access is granted Start: immediately HOW TO APPLY — READ THIS CAREFULLY Applications that do not follow this format are not read. That is deliberate. Following a written spec precisely is the core skill of this job. Start your proposal with this exact block, filled in. Copy the six labels as written, one per line: LANE Timezone: Hours I can commit, in your local time: Agentic tools I have used: Hourly rate: NDA, yes or no: Then answer all seven questions below, numbered, in order. Short answers are fine. We are reading for judgment, not length. Two or three sentences each is plenty for most of them. Do not send a generic portfolio. Do not skip questions. Do not answer them out of order. THE QUESTIONS Which agentic coding tools have you actually used, on what kind of project, and what is the most sessions you have run at the same time? Describe a specific time an AI coding agent proposed something confidently that was wrong. How did you notice, and what did you do? An agent tells you: "The tests are failing because this validation rule conflicts with the new schema. Should I relax the rule or update the migration?" What do you do, and why? An agent says: "I will add a package to handle this. It is a common library and it will save time." What do you do? A session prints the following. What do you report, to whom, and what, if anything, do you do first? FAILED Tests\Feature\Messaging\SendPermitTest -- blocks send without consent Failed asserting that false is true. at app/Services/Consent/ConsentService.php:142 PHPStan: 3 errors app/Services/Messaging/SmsSender.php:88 Parameter #1 $permit of method send() expects SendPermit, null given You are watching 8 sessions. Three of them need attention at the same moment, and a fourth has been idle for twenty minutes. Walk us through your next ten minutes. Name one thing you would refuse to decide on your own in this role, even if you were confident you knew the right answer.