Process Orchestrator — Agent Plugin

Multi-agent chemical process design workflow — from problem statement through requirements, research, concept selection, flowsheet design, equipment sizing, safety review, and gate approval.

Packaged as an Agent Plugins v1.0.0 portable plugin.

What It Does

Given a high-level problem statement (e.g., "Design a compressed air unit for a refinery"), this plugin orchestrates 12 specialized sub-agent skills to produce a complete conceptual design package:

1. Requirements Analysis — Extract structured requirements
2. Innovative Research — Generate 3–6 process concepts
3. Conservative Evaluation — Score concepts for feasibility and risk
4. Concept Detailing — Select and elaborate the winning concept
5. Component List — Identify key chemicals with physical properties
6. Design Basis — Compile Basis of Design document
7. Flowsheet Design — Create process flow narrative
8. Equipment/Stream Catalog — Seed JSON templates
9. Stream Properties — Reconcile heat & material balance
10. Equipment Sizing — Size all equipment (uses PES MCP)
11. Safety Review — HAZOP-style hazard assessment
12. Gate Approval — Final approval memo with financial summary
13. PFD Drawing — Draw the flowsheet as an ISO 10628 PFD in DXF (CAD-editable) + A1 review PDF, with equipment BLOCKs and an extractable equipment schedule (uses pfd-dxf skill + AIBlueprint MCP)

Plugin Structure

process-orchestrator/
├── plugin.json              # Agent Plugins manifest
├── mcp.json                 # PES MCP server config
├── skills/                  # 15 Agent Skills
│   ├── orchestrator/        # Pipeline orchestrator
│   ├── process-requirements-analyst/
│   ├── innovative-researcher/
│   ├── ... (12 sub-agents)
│   ├── project-manager/
│   └── pfd-dxf/             # PFD → DXF + PDF drawing skill
├── vendor/                  # Vendored MCP servers
│   └── aiblueprint-mcp/     # Headless DXF generation (ezdxf + LibreCAD)
├── ai.openclaw/             # OpenClaw extension namespace
│   ├── config/              # LLM defaults, graph config
│   └── hooks/               # Lifecycle hooks
├── scripts/                 # Shared Python scripts
├── references/              # Architecture docs
├── LICENSE
└── CHANGELOG.md

Requirements

• Agent Plugins-compatible client (e.g., OpenClaw, Cursor, VS Code with Agent Plugins support)
• LLM access — OpenRouter or compatible API (configured in ai.openclaw/config/defaults.json)
• PES MCP server — For equipment sizing and stream calculations (configured in mcp.json)
• Python 3.10+ with CoolProp — For physical property lookups

Quick Start

# Clone the plugin
git clone https://github.com/may3rd/process-orchestrator-plugin.git

# Install Python dependencies
cd process-orchestrator-plugin
pip install -r requirements.txt

# Set environment variables
export PES_API_KEY="your-pes-api-key"
export OPENROUTER_API_KEY="your-openrouter-key"

# Run standalone (without client)
python scripts/run_orchestrator.py "Design a compressed air unit for a refinery (300 Nm³/h)"

Configuration

• ai.openclaw/config/defaults.json — LLM provider, models, temperature, tool methods
• ai.openclaw/config/graph.json — Pipeline sequence, state schema, agent reads/writes
• mcp.json — PES MCP server endpoints (Streamable HTTP + stdio fallback) + AIBlueprint MCP (headless DXF)
• vendor/aiblueprint-mcp/ — Vendored AIBlueprint MCP server for CAD/DXF output
• skills/pfd-dxf/ — PFD drawing skill (ISO 10628 DXF + A1 PDF, equipment BLOCKs & schedule)

Quick PFD → DXF

With the design artifacts in hand, draw the PFD from a YAML spec:

python skills/pfd-dxf/scripts/pfd_dxf.py skills/pfd-dxf/examples/distillation.yaml \
  -o out/pfd.dxf --pdf out/pfd.pdf

License

MIT