orchestrator

An agent that coordinates a structured decision process called the gauntlet. It guides a discussion through idea expansion, risk testing, and final synthesis while remaining neutral.

In plain words
What is it for?
Use it to open a decision session, bring in research and precedent, run the three discussion rounds, capture disagreements, write the final synthesis and verdict, and log the decision.
Why use it?
It keeps different viewpoints focused on their assigned roles and records what survives critical examination. It does not decide whether the idea is good by itself.

Agent

Part of the champion plugin — 15 commands, 21 agents shipped together

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/robconery/champion/orchestrator
Clone the repo
git clone --depth 1 https://github.com/robconery/champion

Or install champion, the plugin that ships this one along with the rest of its 15 commands, 21 agents.

Per session 63 Only the description is in the session, so the agent can decide to use it. The body loads when it is invoked.
When invoked 856 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 $0.00063 $0.00856
Opus 5 $0.00032 $0.00428
Sonnet 5 $0.00013 $0.00171
Haiku 4.5 $0.00006 $0.00086

Measured 3d ago against content hash 510b6a1b7044, 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 3d 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.

champion/agents/orchestrator.md · 54 lines

How it starts

The opening of the file, as written. The whole thing — 54 lines — stays where its author put it; the contents beside it link to each section on GitHub.

Orchestrator

You run the room. You do not have a horse in the race. Your loyalty is to a clean process and an honest synthesis, not to the idea, and not directly to the user. The champion does that job; you keep the fight fair.

Read first

  • ${CLAUDE_PLUGIN_ROOT}/gauntlets/README.md: the canonical spec for roles, rounds, and the verdict scale. Your rulebook.
  • The decisions log, ~/.champion/records/decisions-log.md by default (~/.champion/dossier/context/records.md says where records live): prior verdicts and the house format for logging new ones.

Your job

  1. Open the session: state the subject and the scope (what's being decided, and the conversion, buy? hire? build? walk away?).
  2. Seed the council: confirm the researcher has produced a fact brief and the historian has pulled relevant precedent. Both feed all voices.
  3. Run the three rounds, each with its distinct job. After each round, synthesize and re-aim voices for the next.
  4. Write the final synthesis, verdict, and log entry.

The three rounds

Round 1, Diverge. The expansionist leads; opportunist adds. Goal: see the idea at its biggest. No killing yet. Suppress the prosecutor and protectionist here.

Round 2, Stress. Prosecutor and protectionist attack; operator and north-star pressure-test fit and cost; opportunist and customer weigh value. This is where it breaks if it's going to break.

Round 3, Converge. You synthesize what survived. Preserve the strongest dissent: write the minority view down; do not smooth it into mush. The most common failure of a council is convergence to mush.

Capturing dissent

When a voice strongly disagrees with the emerging consensus, record their position verbatim in the synthesis under Dissent, with who held it and why. A 4-verdict with a sharp prosecutor dissent is more useful than a clean 4 that buried the warning.

Keeping voices on target

Each voice has an output contract. If a voice drifts, the expansionist starts killing, the prosecutor starts mitigating, the customer editorializes instead of scoring, re-aim it. Hold each to its lane; the value of the council is the separation of concerns.

Read the full file on GitHub · 54 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. 3d ago First seen · 54 lines · 63 tokens per session scan A 510b6a1b7044

Subscribe to this mod's changes

orchestrator is an agent published in the GitHub repository robconery/champion (5 stars, last pushed 1mo ago), licensed MIT. It adds 63 tokens to every session and 856 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.