loop-runner

What problem does it solve?

Prose-based loops from the Loop Library cannot execute autonomously on their own, requiring a machine contract to enforce bounds, verification steps, and stop conditions to prevent runaway execution and ensure measurable progress.

Core Features & Use Cases

• Enforced 6-Step Loop Cycle: Automatically drives each iteration through Observe → Choose → Act → Verify → Record → Repeat/stop, with hooks that enforce the cycle across conversation turns so loops persist between user messages.
• Dual Verification Modes: Supports machine-verified loops with custom pass/fail commands for reproducible, objective checks, and self-graded loops for fuzzy tasks like documentation refinement, with mandatory evidence requirements for completion to prevent unproven claims of success.
• Configurable Stop Predicates: Lets you set maximum iteration limits, no-progress streaks, and target finish-line commands to ensure loops terminate automatically when goals are met or progress stalls, with a kill switch to disable stop enforcement if needed.
• Use Case Example: If you need to iterate on fixing test failures until the full test suite passes, you can start a loop with a verify command running your test suite and a stop condition for when all tests pass, and the loop will run each cycle, track progress, and stop automatically once the suite is green or the iteration limit is hit.

Quick Start

Use the loop-runner skill to execute a bounded, verifiable Loop Library loop for your repeatable agent workflow by specifying the loop, your verification criteria, and stop conditions.