project-coordinator — Claude Code subagent
project-coordinator is a Claude Code subagent that coordinates mathematics research projects. It discusses goals, formalizes research intent, and delegates work to specialized sub-agents.
no license ★ 240 updated 2026-09-18
.claude/agents/project-coordinator.mdWhat project-coordinator does
project-coordinator serves as the main entry point for an AI co-mathematician research workflow. Rather than proving theorems or writing code itself, it acts as the coordinator between you and three specialized sub-agents: literature-reviewer (paper search), prover (theorem proving), and coder (implementation).
At session start, the subagent loads project state — config, goals, prior decisions, and the registry of active workstreams. If it finds an interrupted background task from a prior session, it surfaces that for your decision. Work follows three phases: first, refining goals through dialogue until you approve; then dispatching workstreams to sub-agents; finally, tracking progress while filtering out low-level execution noise.
Designed for projects containing co-math-config.json and goals.md. Runs multiple workstreams in parallel, with some executing in the background while you interact with the main agent. Interrupted background tasks are automatically detected and surfaced when you resume.
What it helps with
- oversee a multi-stage math research project with parallel workstreams
- lock in research goals before spawning agents, avoiding rework
- track literature review, proof attempts, and code in one place
- resume interrupted work after a session ends
- surface blockers and high-level progress without drowning in execution details
Example prompts for the agent
- Start a new research project on lower bounds for algorithm complexity
- Resume work on a proof and show me where the sub-agents got stuck
- Refine research goals before dispatching to proof and literature search
How to install project-coordinator
Manually: put the files in place
.claude/agents/project-coordinator.mdFor all projects use ~/.claude/agents/. Claude delegates by the description field or an @-mention
Manually: put the files in place
.codex/agents/project-coordinator.tomlCodex defines subagents in TOML: name, description, developer_instructions. Move the Markdown text into developer_instructions by hand. For all projects use ~/.codex/agents/
Manually: put the files in place
.cursor/agents/project-coordinator.mdCursor also reads .claude/agents/
Step by step, with commands for each platform: Claude Code subagents: how to create one, call it and keep context clean
Questions about project-coordinator
How do I install project-coordinator in Claude Code?
Create .claude/agents/project-coordinator.md in your project root. For Codex, use .codex/agents/project-coordinator.toml; for Cursor, use .cursor/agents/project-coordinator.md.
What does project-coordinator do?
It manages a math research project end-to-end. It talks through goals with you, coordinates sub-agents for literature review, proof search, and coding, tracks their progress, and surfaces blockers — without requiring you to monitor low-level execution details.
Can I run multiple workstreams at once?
Yes. The subagent supports parallel workstreams. Some can run in the background while you stay in dialogue with the main agent. Background tasks persist and resume automatically if your session is interrupted.