terminal-bench-loop

What problem does it solve?

Coordinates a bounded, human-in-the-loop process to drive Terminal-Bench problems through Paperclip until a smoke passes or blockers are named. It captures artifacts, diagnoses the exact stop point with /diagnose-why-work-stopped, and routes a board-confirmation for any product fix, then reruns against the same worktree.

Core Features & Use Cases

• Bounded iteration governance: enforces a maximum number of iterations and per-iteration time limits.
• Artifact capture & diagnosis: records run IDs, manifests, results, and the stop reason, and applies /diagnose-why-work-stopped within the loop subtree.
• Isolated worktree discipline: runs against a single Paperclip App worktree with workspace inheritance across iterations.
• Decision routing: gates product fixes behind board confirmation and manages blockers with explicit ownership.

Quick Start

Define the loop inputs (source issue, task name, budget, and workspace), then start the first bounded smoke against the isolated Paperclip App worktree.