Export

/export

Generate Claude Agent SDK Python code from a validated MECE decomposition tree.

Usage

/export 

Examples:

• /export after a /decompose session (exports the last output)
• /export output.json
• /export then paste decomposition JSON

Workflow

1. Verify Validation

The decomposition must pass validation first (overall score >= 0.70). If not validated, I'll run /validate first and show any issues.

2. Generate Code

Using the mapping rules from references/agent_sdk_mapping.md:

• One Agent definition per agent atom
• Orchestration functions per branch type (sequential, parallel, conditional, loop)
• Cross-branch dependency wiring
• Hook-based error handling from atom error modes
• Model tier assignments

3. Output

"""
Agent SDK scaffolding for: [Decomposition Scope]
Dimension: [temporal/functional/etc.]
"""

import asyncio
from agents import Agent, Runner, function_tool

# Node 1.1.1: Step Name
step_name_agent = Agent(
    name="step-name",
    model="claude-sonnet-4-6",
    instructions="""...""",
    tools=[...],
)

# Orchestration
async def execute_phase_1(input_data: str) -> str:
    """Phase 1 (sequential orchestration)"""
    result = input_data
    result = await execute_step_1(result)
    result = await execute_step_2(result)
    return result

async def main(input_data: str) -> str:
    return await execute_root(input_data)

if __name__ == "__main__":
    result = asyncio.run(main("initial input"))
    print(result)

Export Preview (when MCP server connected)

The mece-export-sdk MCP tool renders a preview panel with the generated code and a copy button.

What You Get

• Agent definitions for every agent atom in the tree
• Orchestration functions matching the tree structure
• Comments linking each section to its tree node ID
• A main() entry point that executes the full tree
• TODO markers for human, tool, and external atoms that need implementation

Next Steps

• "Can we refine the prompts for each agent?" -> edit the generated code
• "I need to adjust the decomposition first" -> /decompose or /interview

Source & license

This open-source skill is cataloged on AgentStack and links to its original source — we do not rehost the code.

• Author: fblissjr
• Source: fblissjr/fb-claude-skills
• License: Apache-2.0

Install and usage instructions live in the source repository linked above.