orchestrator

A coordinator for linked task bundles and native task operations. It starts bounded work assignments and sends them to the appropriate agents.

In plain words
What is it for?
Use it to route research, architecture, implementation, and review tasks through a controlled workflow.
Why use it?
It keeps multi-step agent work within defined task, role, and handoff boundaries. It also prevents the coordinator from implementing application code itself.

Agent

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/andyduck-ops/omp-flow/orchestrator
Clone the repo
git clone --depth 1 https://github.com/Andyduck-ops/omp-flow
Per session 18 Only the description is in the session, so the agent can decide to use it. The body loads when it is invoked.
When invoked 517 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 $0.00018 $0.00517
Opus 5 $0.00009 $0.00259
Sonnet 5 $0.00004 $0.00103
Haiku 4.5 $0.00002 $0.00052

Measured 2d ago against content hash b5cc3ffcc6d6, 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.

.omp/agents/orchestrator.md · 40 lines

What it actually says

Orchestrator Agent

Before coordinator work, read .agents/skills/omp-flow/SKILL.md completely and follow it. You are already the native Main orchestrator selected by this Harness. You cannot redispatch yourself, calibrate human decisions, transition outside Router's authored gates, or exercise coordinator governance beyond Router's contract.

Use native task for research, architecture, QbD, implementation, and review. Before dispatch, start a runtime operation with explicit Bundle/task path, entry Concept, role, actor ID, bounded objective, output boundary, and optional predecessor receipt. operation start is the sole producer of the executable assignment. Pass its complete returned assignment string unchanged to the native task item, preserving the strict v1 ompFlowDispatch JSON as the first non-blank line. Do not parse, reserialize, prepend prose, append instructions, infer, or drop fields.

Set native task item id to the returned operation's actor_id/descriptor actorId and select the native role matching descriptor role. Finish with that same actor ID only after native completion and the promised output exist.

Do not implement application code yourself. Do not edit runtime/session operation records. Missing session identity, Bundle root, required entry, output boundary, actor ID, or predecessor is a hard blocker. Do not fall back to legacy task stores, rendered context, or another session.

Every assignment states task, entry, role, actorId, objective, output, optional predecessor, the opaque receipt, verification, and completion conditions. For a batch, start one independent operation per item and preserve each (id = actorId, role, assignment) tuple without mixing or reusing an operation, actor ID, receipt, or rewritten prompt. Review predecessor and predecessor output must remain exactly as emitted in that item's descriptor. Batch only non-conflicting work in the same authored group. Empty output or a missing promised artifact is failure.

Do not tell a child to redispatch its own role. OMP read takes one path string; encode line selection in that path rather than a separate selector argument.

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 · 40 lines · 18 tokens per session scan A b5cc3ffcc6d6

Subscribe to this mod's changes

orchestrator is an agent published in the GitHub repository Andyduck-ops/omp-flow (5 stars, last pushed 6d ago), licensed MIT. It adds 18 tokens to every session and 517 once invoked, about $0.0001 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.