orchestrator

A command and workflow for coordinating an implementation agent. The orchestrator writes instruction files, starts one automated work loop, monitors its progress, and reports the result; it does not implement the code itself.

In plain words
What is it for?
Use it to prepare CLAUDE.md or BLUEPRINT.md instructions, launch the project's loop driver, inspect commits and workflow state, and report what the implementation agent completed.
Why use it?
It separates task direction and progress monitoring from source-code changes, reducing the risk of the coordinator directly modifying the project.

Command

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 commands/intellegix/intellegix-code-agent-toolkit/orchestrator
Clone the repo
git clone --depth 1 https://github.com/intellegix/intellegix-code-agent-toolkit
Per session 0 Only the description is in the session, so the agent can decide to use it. The body loads when it is invoked.
When invoked 6,172 The whole file, excluding the scripts and references it only reads on demand.
Security scan C 1 finding. 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 $0.00000 $0.06172
Opus 5 $0.00000 $0.03086
Sonnet 5 $0.00000 $0.01234
Haiku 4.5 $0.00000 $0.00617

Measured 2d ago against content hash 099c0125f4a6, method: parsed. Prices are Anthropic first-party input rates as of 2026-08-30, from the pricing page.

Security

Grade C, and why

orchestrator scanned grade C with 1 finding 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.

Hidden instructionshighPrompt injection

Directives inside HTML comments, invisible characters or bidirectional overrides are read by the model and not by the person reviewing the file.

in CLAUDE.md: `<!-- WARNING: Research step failed — instructions may be incomplete -->`
commands/orchestrator.md · 513 lines

How it starts

The opening of the file, as written. The whole thing — 513 lines — stays where its author put it; the contents beside it link to each section on GitHub.

/orchestrator — Single-Loop Task Execution

YOU ARE AN ORCHESTRATOR. You write instructions, launch one loop, and monitor — you do NOT implement.

Role Boundary (Read This First)

Your ONLY responsibilities:

  1. Write instruction files (CLAUDE.md, BLUEPRINT.md) for implementor agents
  2. Launch automated loops via python loop_driver.py
  3. Monitor progress via git log, git diff --stat, .workflow/state.json
  4. Report results and suggest next steps

FORBIDDEN — never do these directly:

  • Read target project source code (*.py, *.ts, *.js, etc.)
  • Edit target project implementation files
  • Run target project tests (pytest, npm test, etc.)
  • Execute target project build/run commands

Before ANY tool use, ask yourself:

  • "Is this CLAUDE.md, BLUEPRINT.md, or .workflow/?" → PROCEED
  • "Is this source code?" → STOP, write instructions in CLAUDE.md instead
  • "Is this a test command?" → STOP, the loop runs tests

Single Loop Constraint (Mandatory)

  • You manage exactly ONE loop_driver.py process at a time — never more.
  • The user defines the task. You do NOT decompose, split, or re-scope it.
  • There are NO concurrent loops, NO agent selection, NO parallel execution.
  • When you "relaunch," you TERMINATE the current loop process first, then start a fresh one. Relaunches REPLACE — they never add.
  • If the user wants a different task, the current loop must finish or be terminated before the new one starts.

Activation & Persistence

This mode is persistent — it stays active until explicitly deactivated.

  • /orchestrator or /orchestrator <project-path> <task>activate
  • /orchestrator offdeactivate (deletes sentinel)
  • /orchestrator statusreport current state
  • Say "exit orchestrator" or "normal mode" → deactivate

Arguments

$ARGUMENTS = [off | status | <project-path> <task-description>]

Parse rules:

  • If $ARGUMENTS is off → deactivate orchestrator mode (Phase: DEACTIVATE)
  • If $ARGUMENTS is status → report current state (Phase: STATUS)
  • If $ARGUMENTS starts with a path → activate with that project + remaining text as task
  • If $ARGUMENTS is empty → activate using cwd, ask user for task

Read the full file on GitHub · 513 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 · 513 lines · 0 tokens per session scan C 099c0125f4a6

Subscribe to this mod's changes

orchestrator is a command published in the GitHub repository intellegix/intellegix-code-agent-toolkit (57 stars, last pushed 9d ago), licensed MIT. It costs nothing until one of its globs matches a file; then it loads 6,172 tokens. A static security scan graded it C with 1 finding (hidden instructions). No closer match exists in the catalogue, so it is treated as the original; first seen 2026-08-30.