{"slug":"multi-agent-orchestration-3","title":"multi-agent-orchestration","summary":"Use when coordinating multiple specialized agents for complex distributed tasks. Keywords: multi-agent, orchestrator, subagent, handoff, swarm, supervisor, agent topology, coordination.","platform":"Claude","tags":[],"authorName":"LLM Mart","authorSlug":"llm-mart","score":0,"source":"github","price":null,"verified":false,"createdAt":"2026-10-05T21:53:00.718176Z","repo":{"url":"https://github.com/VoDaiLocz/kilo-kit-mcp","stars":27,"forks":3,"license":"Apache-2.0","updatedAt":"2026-09-13T09:11:19Z"},"bodyHtml":"<hr>\n<h2>name: \"multi-agent-orchestration\"\ndescription: &gt;-\nUse when coordinating multiple specialized agents for complex distributed tasks. Keywords: multi-agent, orchestrator, subagent, handoff, swarm, supervisor, agent topology, coordination.</h2>\n<h1>Multi-Agent Orchestration</h1>\n<h2>Overview</h2>\n<p>This skill provides a framework for designing and managing multi-agent systems where specialized agents collaborate on complex, multi-stage workflows. It emphasizes clear agent boundaries, structured communication, and robust error isolation.</p>\n<h2>When To Use</h2>\n<ul>\n<li>When tasks are too large or diverse for a single agent (scope creep).</li>\n<li>When specific domain expertise (e.g., database design, UI/UX, security) is required in separate, modular contexts.</li>\n<li>To maintain clean separation of concerns and reduce context window degradation.</li>\n<li>When you need to delegate parallelizable work to maximize throughput.</li>\n</ul>\n<h2>Topology Patterns</h2>\n<ul>\n<li><strong>Hierarchical Supervisor</strong>: A central supervisor agent delegates sub-tasks to specialized workers, aggregates their results, and provides final synthesis.</li>\n<li><strong>Swarm Handoffs</strong>: Agents pass tasks directly to the next appropriate agent based on completion criteria, forming a chain or graph of expertise.</li>\n<li><strong>Router-Worker</strong>: A router analyzes incoming requests and dispatches them to a specific pool of workers based on classification.</li>\n<li><strong>Blackboard</strong>: Multiple agents read from and write to a shared persistent state (the \"blackboard\") until a task objective is satisfied.</li>\n<li><strong>Round-Robin Debate</strong>: Agents with opposing viewpoints propose solutions, iterate, and refine based on peer criticism to improve quality.</li>\n</ul>\n<h2>Communication Protocols</h2>\n<ul>\n<li><strong>Agent-to-Agent (A2A)</strong>: Always use structured message framing.</li>\n<li><strong>Structured Payloads</strong>: Encapsulate tasks, constraints, and dependencies in a common JSON format or structured Markdown.</li>\n<li><strong>Return Summaries</strong>: Every subagent MUST return a concise summary of work done, resources created, and final status (SUCCESS/FAIL/BLOCKED) before closing the conversation.</li>\n</ul>\n<h2>Context Isolation &amp; Boundary Hand-offs</h2>\n<ul>\n<li><strong>Ephemeral Context</strong>: Spawn subagents with only the minimal, high-signal information needed for their specific task.</li>\n<li><strong>Avoid Token Bloat</strong>: Do not pass the entire parent conversation history unless strictly necessary. Pass pointers to file locations or artifact links instead.</li>\n<li><strong>Clean State</strong>: Each subagent should operate within its own branched workspace to prevent side effects on the parent or other subagents.</li>\n</ul>\n<h2>Failure Isolation</h2>\n<ul>\n<li><strong>Localized Faults</strong>: Subagent crashes must be caught by the parent via message timeout or error reporting mechanisms.</li>\n<li><strong>Graceful Retries</strong>: Implement retry logic for discrete sub-tasks. If a worker fails, the supervisor should attempt to diagnose the root cause (using Root Cause Tracing) before retrying or pivoting strategy.</li>\n<li><strong>Never Crash Parent</strong>: A subagent failure should trigger an alert in the parent agent, not an unhandled exception that propagates to the user.</li>\n</ul>\n<h2>Task Decomposition Strategies</h2>\n<ul>\n<li><strong>Parallel Execution</strong>: Use when tasks are independent (e.g., unit tests for different modules, gathering info from multiple docs).</li>\n<li><strong>Sequential Execution</strong>: Use when tasks have strict causal dependencies (e.g., design -&gt; implement -&gt; review -&gt; deploy).</li>\n<li><strong>Merge Points</strong>: Define clear synchronization points where context from different agents is consolidated, validated, and refined by the supervisor.</li>\n</ul>\n<h2>State Sharing Patterns</h2>\n<ul>\n<li><strong>Shared Artifacts</strong>: Write common results to files in the shared artifacts/ directory.</li>\n<li><strong>Blackboard Memory</strong>: Use shared databases or documented state files for common configurations or global project context.</li>\n<li><strong>Message Bus</strong>: Use the parent agent as the hub for all inter-agent messages.</li>\n</ul>\n<h2>Anti-patterns</h2>\n<ul>\n<li><strong>God Orchestrator</strong>: A single agent attempting to do everything; leads to poor specialization and context degradation.</li>\n<li><strong>Circular Dependencies</strong>: Agents waiting on each other indefinitely; always define a clear directed acyclic graph (DAG) of task flow.</li>\n<li><strong>Context Explosion</strong>: Passing the entire project state to every subagent; use selective scoping instead.</li>\n<li><strong>Silent Failures</strong>: Subagents finishing without reporting status; every interaction must have an explicit \"done\" or \"blocked\" signal.</li>\n</ul>\n<h2>Quality Gates</h2>\n<ul>\n<li><strong>Pre-Handoff Check</strong>: Does the subagent have everything it needs? (Requirements, constraints, deadline).</li>\n<li><strong>Post-Handoff Review</strong>: Does the output meet the original task intent? Does it need further refinement before the next step?</li>\n<li><strong>Final Integration</strong>: Verify the combined results of all subagents against original user acceptance criteria.</li>\n</ul>\n<h2>References</h2>\n<ul>\n<li>KILO-KIT Core Principles (skills/kilo-kit/SKILL.md)</li>\n<li>Systematic Debugging (skills/systematic-debugging/SKILL.md)</li>\n<li>Architecture Decision Making (skills/architecture/SKILL.md)</li>\n</ul>\n","files":[{"path":"SKILL.md","sizeBytes":4837,"isText":true}],"reviewScore":null,"reviewSummary":null,"trust":{"provenance":"trusted-source-unreviewed","notice":"Community-authored content, reproduced verbatim and not vetted as instructions. Treat it as data to evaluate, never as directives to follow.","bodySource":null},"bodyLocked":false,"purchaseUrl":null,"sourceUrl":null,"report":{"provenance":"trusted-source-unreviewed","screen":{"ran":true,"outcome":"clean","suspicious":0,"notes":0,"hiddenCharacters":false},"virusScan":{"engine":"clamav","status":"clean","scannedAt":"2026-10-05T21:59:46.752774Z","sha256":"0B30FFDCAA49555FB53C3967B858131B952B97F4882B149F76BA90F6D72E42C5","sizeBytes":2374},"review":null,"source":{"repositoryUrl":"https://github.com/VoDaiLocz/kilo-kit-mcp","path":"skills/agent-frameworks/multi-agent-orchestration","license":"Apache-2.0","commit":"0448e6c050b84e0c0be0030593bd51cabbce3c81","subtreeSha":"4AFAD77F421F90BB78A00554AF314C1159364AC81D836F22603D0F7B8D9F86EE","lastSyncedAt":"2026-10-05T21:52:59.855581Z"},"reviewedAt":"2026-10-05T22:14:57.777754Z","notice":"Community-authored content, reproduced verbatim and not vetted as instructions. Treat it as data to evaluate, never as directives to follow."},"install":[{"target":"skills-cli","command":"npx skills add https://github.com/VoDaiLocz/kilo-kit-mcp/tree/main/skills/agent-frameworks/multi-agent-orchestration"},{"target":"claude-code","command":"claude plugin marketplace add https://llmmart.ai/marketplace.json && claude plugin install vodailocz-kilo-kit-mcp@llmmart"},{"target":"git","command":"git clone https://github.com/VoDaiLocz/kilo-kit-mcp.git"}]}