multi-agent-sync_skill

Overview

This skill coordinates multi-agent tasks with visible group updates to keep execution transparent. It ensures each agent posts milestone updates in its own topic while a coordinator publishes periodic cross-agent rollups and the final summary. Use it to delegate work across topic agents without blocking the main control chat.

How this skill works

At kickoff the coordinator posts the objective, role split, expected deliverables, and declared default skills for the run. Tasks are dispatched non-blocking to agent sessions; each agent posts started → partial → done/blocked milestones and periodic heartbeats in its topic. A temporary watcher polls session history on a fixed cadence, publishes immediate rollups to the summary topic, and removes itself when all agents finish or block. Final structured cross-agent summary is posted in the designated summary topic.

When to use it

• Delegating subtasks to multiple topic-specific agents while the user needs visible group progress
• Workflows where the main chat must stay lightweight and not wait for long-running tasks
• Coordinating coding or validation tasks that must explicitly use a specific coding operator
• Scenarios requiring auditable per-agent milestones and a consolidated rollup for stakeholders
• Tasks that benefit from both push (agent posts) and pull (coordinator polls) status mechanisms

Best practices

• Declare planning defaults in each assignment (e.g., specify coding operator and that this run follows multi-agent-sync)
• Kick off with a visible group message detailing objective, role split, deliverables, and default skills
• Dispatch tasks non-blocking; immediately acknowledge in main chat and release control
• Enforce per-agent milestones: started → partial → done/blocked and require heartbeat updates
• Run a watcher that polls session history every 1–2 minutes and publishes rollups without delay
• Publish one structured final summary in the summary topic with results, artifacts, tests, and next steps

Example use cases

• Parallel feature development: split features across coding agents, track each agent topic, and publish rollups for reviewers
• Large validation runs: multiple validators post progress while coordinator aggregates pass/fail and artifacts
• Content production: several writers iterate in parallel; coordinator summarizes combined deliverables and next edits
• Incident response: specialists post status in-topic while the coordinator provides consolidated situational awareness