agentleFS
Sign inStart free
asmamestasmamest/Nexus-Skills/skills/agentic-engineering/multi-agent-orchestration/SKILL.md

multi-agent-orchestration

Skill1 starsChanged 17 days agoMITView on GitHub

Design and evaluate supervisor, router, handoff, and parallel-worker agent patterns with clear ownership, bounded delegation, structured messages, and conflict resolution.

We split this skill into the rules it gives an agent

5 rules. 0 are common, 5 are distinctive, and 0 are very specific to this project.

  • 0

    Common

    Most of the 14 edited versions we read keep these.

  • 5

    Distinctive

    This author's own choices. Worth a look.

  • 0

    Very specific

    Tied to their project's names, files or tools. Probably skip.

  • Common
  • Distinctive
  • Very specific
  1. Distinctive:

    Use multiple agents only when specialization, parallelism, or isolation creates measurable value.

    line 9
  2. Distinctive:

    Choose orchestration: Use a router for classification, a supervisor for dynamic delegation, explicit handoffs for control transfer, or …

    line 14
  3. Distinctive:

    Keep sequential dependencies explicit.

    line 25
  4. Distinctive:

    Use a reviewer agent only when it has clear criteria and can inspect the actual artifact/evidence.

    line 26
  5. Distinctive:

    Avoid majority voting among agents that share the same model, prompt, or source bias as a substitute for verification.

    line 27

“Common” is measured: kept by at least 6 in 10 of the 14 edited versions of this file we could read. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.

Keep what fits and drop the rest. Your version, ready for your agents.
Files
SKILL.md

Multi-Agent Orchestration

Goal

Use multiple agents only when specialization, parallelism, or isolation creates measurable value. Make delegation bounded, auditable, and recoverable.

Workflow

  1. Prove the need. Compare a single-agent or deterministic baseline with the proposed multi-agent design.
  2. Define roles. Each agent needs a narrow responsibility, input contract, output contract, tool allowlist, and stop condition.
  3. Choose orchestration. Use a router for classification, a supervisor for dynamic delegation, explicit handoffs for control transfer, or parallel workers for independent tasks.
  4. Bound delegation. Set maximum depth, number of agents, wall time, token/cost budget, retries, and tool calls. Prevent recursive spawning loops.
  5. Structure messages. Include task objective, relevant context, constraints, evidence references, expected output schema, and completion criteria. Avoid passing entire conversation histories by default.
  6. Isolate privileges. Give each agent only the data and tools required. Never assume an agent's identity or assertions establish authorization.
  7. Handle parallel results. Define merge rules, deduplication, conflict detection, source precedence, and what happens when workers disagree or fail.
  8. Verify outputs. Validate specialist results against source evidence and task requirements before taking action or presenting conclusions.
  9. Observe the delegation graph. Trace parent/child calls, costs, failures, retries, and decisions.
  10. Evaluate. Measure end-to-end success, latency, cost, duplicated work, disagreement rate, and policy violations against the baseline.

Coordination patterns

  • Parallelize independent, bounded research or analysis.
  • Keep sequential dependencies explicit.
  • Use a reviewer agent only when it has clear criteria and can inspect the actual artifact/evidence.
  • Avoid majority voting among agents that share the same model, prompt, or source bias as a substitute for verification.

Failure handling

Define behavior for worker timeout, malformed output, partial results, contradictory claims, supervisor failure, cancellation, and budget exhaustion. Preserve completed work when safe, but do not repeat external side effects blindly.

Deliverable

Provide a delegation diagram, role contracts, state transitions, budget limits, permissions, merge/conflict policy, trace fields, and comparative evaluation.

References

Download now (ZIP)

Does it work?

Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.

No reviews yet

People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.

  • It worked
  • Partly
  • It didn't
What did you use it for? e.g. Reviewing pull requests in a Rails app

Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.

Your trail

Files you open will collect here, so you can find them again.

Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.

Now do this for your own.

Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.

Check your setup, free