Agent In the Loop

Agent in the Loop — Orchestrate. Verify. Ship. (controsl and runs vscode isolated extensions and controls E2E) Tired of “Fixed! ✅” that isn’t? Agent in the Loop is a control-room for autonomous coding agents that kills false positives and surfaces only work that’s actually done. We launch isolated VS Code environments in parallel, pairing Cline (implementer) with CodeRabbit (reviewer). A referee routes messages via MCP tools—cline, coderabbit, claude-code—and accepts one truth source: make verify. If it isn’t green, it isn’t real. Why it works: Parallel orchestration: multiple agent instances attack the task from different angles. Independent QA: claude-code runs make verify in a clean env—no excuses. Evidence-first UI: QA chips, diff preview, and a concise trail of agent messages. Reviewer in the loop: rework is automatic; you only review when it’s verified. Stop babysitting tests. Start reviewing results. Turn 0.5× days back into 10× output.