reza

tools

Conductor

A multi-agent orchestration harness for Claude Code. Work flows down, only verified claims flow up, and hooks record every action.

Source on GitHub →

Install

Requires Claude Code, python3 3.9+ and git. No other dependencies.

bash
claude plugin marketplace add reza-phys/conductor
claude plugin install conductor@conductor

Then, in a project you want it for:

text
/conductor:init     creates .conductor/ (config, standards, plans)
/conductor:serve    live dashboard on http://127.0.0.1:8765

Ask for the work as you normally would and the session plans it, dispatches agents and comes back to you when a decision is yours. /conductor:status shows who is doing what, at any time, for no tokens. Conductor only activates in projects that contain .conductor/; everywhere else its hooks exit silently.

How it works

Conductor architecture: work flows down, only verified claims flow up

Every result passes an independent audit gate before it reaches you.

Conductor internals: only the hook writes the log; every view replays it

One append-only log, written only by the hooks, so the dashboard and the provenance trail never depend on an agent remembering to report.

Why it works this way, and where it can still be wrong: the post.