todokan-review-loop_skill

Overview

This skill processes Todokan task and thought boards using a review-loop workflow that autonomously handles items in the doing state. It reads task context and full comment threads, crafts a concise context-aware MCP reply, posts an execution update, and moves the task to done (Review) when the objective is addressed or a concrete blocker is posted. It is designed for recurring polling/cron automation with Todokan MCP.

How this skill works

The agent enumerates habitats and boards, finds tasks with status doing, and ensures full task fields (title, description, labels, dueDate, priority). For each task it reconstructs a strict timeline of comments, identifies the latest user intent or active unresolved question, and decides whether to answer directly or spawn a single internal research subagent. It then composes a short, objective-aligned MCP comment, optionally attaches a document for large outputs, and updates status to done only when the task objective is met or a blocking question is posted.

When to use it

• Automate periodic pickup of tasks that enter doing and need a reply or closure.
• Ensure consistent single-voice KAM responses across many Todokan boards.
• Handle follow-up questions, small clarifications, or quick execution updates.
• Run as cron/polling automation tied to Todokan MCP for recurring workflows.
• Prevent stale doing items by enforcing a clear review-and-complete cycle.

Best practices

• Always reconstruct title+description into one sentence objective before replying.
• Read the full comment thread and treat latest unresolved user question as primary.
• Answer directly in the first line when a clear factual or direct question exists.
• Spawn exactly one research subagent only when factual certainty or cross-item synthesis is required.
• Attach a document and post a short summary if the response exceeds ~600 characters or needs structured content.

Example use cases

• A task moved to doing with a customer question: read thread, answer the question, update progress, and close to done.
• Periodic cron run that finds doing tasks with missing descriptions, recovers context, asks one clarifying question, and leaves the task doing if unresolved.
• Cross-board check: detect related tasks that affect the objective, run a short research subagent, summarize findings, and mark done with evidence.
• Acknowledgement flows where the latest user comment is just thanks: post a brief confirmatory reply or skip posting to avoid repetition.