orchestrator

An agent coordinator that assigns parts of a software task to specialist agents such as architects, implementers, and testers. It uses task type to choose the right worker.

In plain words
What is it for?
Use it to route architecture reviews, implementation work, and quality checks to different agents. It is also useful for testing how agent relationships are declared and discovered.
Why use it?
It removes the need to manually decide which specialist should handle each part of a task. It also keeps planning, implementation, and checking as separate responsibilities.

Agent for Claude Code

Install

Getting it into your agent

One page per mod, every tool's command on it. A separate URL per tool would split the same page into five that compete with each other.

agentmods
npx agentmods add agents/configflux/weld/orchestrator
Clone the repo
git clone --depth 1 https://github.com/configflux/weld

Made for: Claude Code.

Per session 67 Only the description is in the session, so the agent can decide to use it. The body loads when it is invoked.
When invoked 640 The whole file, excluding the scripts and references it only reads on demand.
Security scan A 0 findings. Scan, not verified.
Origin original No closer match found in the catalogue.
Token cost

What it costs to keep this loaded

Counted locally with the o200k_base tokenizer, which is exact for GPT models; Claude uses its own tokenizer and its counts differ. Treat this as one consistent yardstick across the catalogue rather than a bill. Prices are per million input tokens.

ModelPer sessionOnce invoked
Fable 5 $0.00067 $0.00640
Opus 5 $0.00034 $0.00320
Sonnet 5 $0.00013 $0.00128
Haiku 4.5 $0.00007 $0.00064

Measured 2d ago against content hash d79cd44bf809, method: parsed. Prices are Anthropic first-party input rates as of 2026-08-30, from the pricing page.

Security

Grade A, and why

orchestrator scanned grade A with 0 findings against 26 rules in 11 categories — prompt injection, anti-refusal, data exfiltration, privilege escalation, supply chain, agent snooping, system-prompt leakage, SSRF and excessive agency — measured 2d ago.

A static scan of the body, not an audit. Every finding is printed with the line that produced it so you can judge whether it matters here. A mod is markdown that instructs an agent; that is exactly why what it instructs is worth reading.

Nothing flagged

None of the 26 patterns this scan looks for appear in this file: no shell pipes, no recursive deletes, no credential paths, no hidden text, no instruction-override or anti-refusal phrasing, no agent-config snooping. That is not a guarantee, it is the absence of the things that are checkable.

examples/agent-graph-demo/.claude/agents/orchestrator.md · 97 lines

How it starts

The opening of the file, as written. The whole thing — 97 lines — stays where its author put it; the contents beside it link to each section on GitHub.

Orchestrator

This agent is the demo's worker-shaped orchestrator. Every literal subagent_type: reference below is a slice-1 empirical pattern (the q8rl regex). The single <implementer_type> placeholder is the slice-2 empirical pattern (frontmatter declaration via weld: invokes_agents: resolves it without polluting the inferred-edge regex).

The boundary vs. existing demo agents:

  • agent:planner produces the plan; the orchestrator consumes it.
  • agent:reviewer is a leaf reviewer the orchestrator dispatches to.
  • The orchestrator itself never edits source -- it routes to specialists.

Phase 1 -- Architecture review (literal references)

Agent(
  name: "architect",
  subagent_type: "architect",
  prompt: "Write an ADR if the task class is ARCHITECTURE."
)
Agent(
  name: "qa",
  subagent_type: "qa",
  prompt: "Verify acceptance criteria once the implementer reports done."
)

Phase 2 -- Implementer dispatch (placeholder indirection)

The implementer is selected by task_class. The literal name is not known at graph-discovery time -- declaring the candidate set in frontmatter (weld: invokes_agents) is the only authoritative way to surface those edges.

Agent(
  name: "dev",
  subagent_type: "<implementer_type>",
  prompt: "Implement bd-<id>. Run the gate before returning."
)

Phase 3 -- Specialist review chain (literal references)

Agent(
  name: "reviewer",
  subagent_type: "reviewer",
  prompt: "4-eye review of the dev diff. Findings only -- no edits."
)
Agent(
  name: "build-fixer",
  subagent_type: "build-fixer",
  prompt: "Fall through to the build-fixer if Bazel breaks."
)
Agent(
  name: "security",
  subagent_type: "security",
  prompt: "Run security screen if required."
)

Phase 4 -- Hand-off

The orchestrator delegates final acceptance to the reviewer chain. See @docs/architecture/principles.md for the underlying authority rule and skill:architecture-decision when an ADR is required. Use mcp:filesystem for any file-context lookups.

Read the full file on GitHub · 97 lines

Changes

What this file has done since we first saw it

Hashed on every crawl. A supply-chain change to an agent config is a question of when, not whether, so the history is kept rather than the latest state alone.

  1. 2d ago First seen · 97 lines · 67 tokens per session scan A d79cd44bf809

Subscribe to this mod's changes

orchestrator is an agent published in the GitHub repository configflux/weld (6 stars, last pushed 3d ago), licensed Apache-2.0. It adds 67 tokens to every session and 640 once invoked, about $0.0003 per session on Opus 5. A static security scan graded it A with 0 findings. No closer match exists in the catalogue, so it is treated as the original; first seen 2026-08-31.