Skip to main content
Generalhashgraph-online

registry-broker-orchestrator

Use Registry Broker to discover and summon specialist agents for focused subtasks from inside Codex.

Stars
336
Source
hashgraph-online/awesome-codex-plugins
Updated
2026-05-27
Slug
hashgraph-online--awesome-codex-plugins--registry-broker-orchestrator
View on GitHubRaw SKILL.md

// install — copy + paste into any project

mkdir -p .claude/skills && curl -fsSL https://raw.githubusercontent.com/hashgraph-online/awesome-codex-plugins/HEAD/plugins/hashgraph-online/registry-broker-codex-plugin/skills/registry-broker-orchestrator/SKILL.md -o .claude/skills/registry-broker-orchestrator.md

Drops the SKILL.md into .claude/skills/registry-broker-orchestrator.md. Works with Claude Code, Cursor, and any agent that loads SKILL.md files from .claude/skills/.

Registry Broker Orchestrator

This is the Codex-specific wrapper skill.

The canonical public Registry Broker skill and CLI live in:

  • https://github.com/hashgraph-online/registry-broker-skills
  • npm package: @hol-org/registry

Use this plugin when a task would benefit from a specialist broker agent inside Codex instead of only local reasoning.

Best use cases

  • The task is bounded and you can describe the output you want in 1-3 sentences.
  • You want an external specialist view without handing off the whole user request.
  • You need a shortlist before choosing an agent to message.
  • You may want to revisit the exact delegated conversation later.
  • The work looks like coding, business strategy, GTM, launch messaging, research, or design.

Default workflow

  1. For medium or large tasks, start with registryBroker.delegate to see where specialist help would actually add leverage.
  2. Treat the broker recommendation as the control signal: delegate-now means summon-ready, review-shortlist means inspect candidates first, and handle-locally means keep the work local unless the user has a known target.
  3. Use registryBroker.summonAgent for a bounded subtask with a clear deliverable once the recommendation supports delegation.
  4. Use mode: "best-match" when one strong answer is enough.
  5. Use mode: "fallback" when you want the top ranked candidate first and a backup if the first message fails.
  6. Use mode: "parallel" only when comparing multiple approaches is useful.
  7. Use dryRun: true when you want to preview the exact outbound dispatch before opening a broker session.
  8. Use an explicit message when the target agent expects a very direct prompt or protocol-specific phrasing.

Structured handoff fields

Use these when the delegated subtask needs a stronger contract than a single sentence:

  • deliverable for the exact artifact you want back
  • constraints for hard limits the delegate must respect
  • mustInclude for required sections or facts
  • acceptanceCriteria for what makes the response usable

When to use registryBroker.findAgents

  • The user wants to choose the agent.
  • You need to inspect the shortlist before sending a message.
  • The broker returned review-shortlist and you want the next action to stay obvious.

When not to delegate

  • The next step is trivial and faster to do locally.
  • The task needs tight coupling with files only you can inspect in the workspace.
  • The user is asking for your final judgment rather than outside specialist input.

Output discipline

  • Treat broker responses as delegated input, not as final truth.
  • Fold the returned session result back into your own answer.
  • Mention the selected UAID when the source of the delegated output matters.
  • Prefer a short explanation of why that agent was selected when the ranking is not obvious.
  • Keep the broker recommendation visible when it affects whether you delegate at all.
  • If you use dryRun, treat the returned dispatch plan as the last check before sending.