Scaffold

The scaffold MCP tool downloads role-specific files into the calling agent's workspace. After onboarding (which creates the folder structure), scaffold gives the agent its AGENTS.md and the guides it needs for its role.

How it works

1. Agent connects to Conductor MCP
2. Agent calls onboard → creates folder structure, clones conductor
3. Agent calls scaffold("engineering-lead") → downloads role files
4. Agent reads conductor/AGENTS.md → knows what it can/cannot do
5. Agent starts working

Role files

Each role gets:

Role	AGENTS.md	Guides
Portfolio Manager	conductor/AGENTS.md	workflow, portfolio, lines/index, gatekeepers, notion-hierarchy, initiatives
Product Manager	conductor/AGENTS.md	product, estate/chooser, gatekeepers
Engineering Lead	conductor/AGENTS.md	software/overview, patterns/, lifecycle/, estate/*, secrets, github-first, socket-activation, gatekeepers
Editor-in-Chief	conductor/AGENTS.md	editorial, lines/index, gatekeepers
Engagement Manager	conductor/AGENTS.md	services, privacy-model, gatekeepers
Platform Engineer	conductor/AGENTS.md	operate, socket-activation, deploying, gatekeepers

All roles also get about/product.md and about/organization.md as shared context.

The AGENTS.md

Each role's AGENTS.md tells the agent: - What tools it needs - What it can do - What it cannot do - What guides to read - Key rules and non-negotiables

The AGENTS.md is the single source of truth for the agent's behavior. It replaces ad-hoc prompting with structured instructions.

Implementation

The tool returns JSON with file paths and their contents:

{
  "role": "engineering-lead",
  "files": {
    "conductor/AGENTS.md": "# AGENTS.md — Engineering Lead\n\n...",
    "conductor/docs/lines/software/overview.md": "# Software line\n\n...",
    ...
  },
  "instruction": "Write each file to disk at the path shown..."
}

The agent writes each file to disk, then reads its AGENTS.md to begin work.