iterating-to-completion

What problem does it solve?

This skill prevents two failure modes: premature exit and infinite loops in AI task execution, ensuring progress and guardrails.

Core Features & Use Cases

• Completion Promises: explicit completion signals per role to mark success. The agent outputs this string ONLY when criteria are truly met, e.g., architecture plan completion, implementation completion, review completion, test plan completion, and all tests passing.
• Scratchpad Protocol: maintain a per-task scratchpad at .jonggrang/.ephemeral/scratchpad-{task_id}.md with sections for Completed Steps, Current Focus, Failures & Learnings, and Next Step.
• Loop Detection: monitor the last outputs to detect stagnation or repetition; trigger strategy changes or escalation when similarity exceeds thresholds.

Quick Start

Guide the agent to iterate tasks until all completion signals are observed, using the completion tokens and the scratchpad to manage progress.