Use when planning complex Speckle workflows, choosing the right connector for a task, or coordinating multi-tool data exchange. Prevents incorrect connector selection, terminology confusion (Stream vs Project), and broken cross-tool federation workflows. Covers intelligent workflow orchestration, connector selection decision trees, send/receive sequence planning, cross-tool federation coordination, terminology resolution, version strategy recommendations, and multi-model coordination. Keywords: speckle workflow, orchestrate, coordinate, connector selection, federation, multi-tool, cross-tool, planning, strategy, which connector, how to combine tools, exchange data.
Install with Codex or Claude Copy this prompt, paste it into Codex, Claude, or another assistant, and let it review the skill page and install it for you.
A direct command skips the review prompt. Inspect the source before running it.
Use when planning complex Speckle workflows, choosing the right connector for a task, or coordinating multi-tool data exchange. Prevents incorrect connector selection, terminology confusion (Stream vs Project), and broken cross-tool federation workflows. Covers intelligent workflow orchestration, connector selection decision trees, send/receive sequence planning, cross-tool federation coordination, terminology resolution, version strategy recommendations, and multi-model coordination. Keywords: speckle workflow, orchestrate, coordinate, connector selection, federation, multi-tool, cross-tool, planning, strategy, which connector, how to combine tools, exchange data.
license
MIT
compatibility
Designed for Claude Code. Requires Speckle Server 2.x/3.x, any Speckle Connector.
metadata
{"author":"OpenAEC-Foundation","version":"1.0"}
speckle-agents-model-coordinator
This is the META-AGENT skill. It references ALL other skills in the package and helps Claude orchestrate complex Speckle workflows by selecting the right skills, connectors, and strategies.
Skill Package Map
Use this map to determine which skill to consult for a given task.
Task Domain
Skill
When to Use
Object model basics
speckle-core-object-model
Understanding Base objects, serialization, id vs applicationId
GraphQL/REST API
speckle-core-api
Authentication, API calls, terminology mapping
Transport layer
speckle-core-transport
Sending/receiving objects, ServerTransport, local cache
ALWAYS use current terminology. Resolve legacy terms immediately when encountered.
User Says
Correct Term
API Parameter Name
Notes
Stream
Project
stream_id / projectId
SDK params still use stream_id
Branch
Model
branch_name / modelId
Legacy branch_name in SpecklePy
Commit
Version
commit_id / versionId
GraphQL uses both schemas
Collaborator
Project Collaborator
varies
Role-based access
When a user mentions "stream", ALWAYS translate to "project" in your response and note the mapping. When writing code, use the parameter name the SDK expects (which may still be stream_id).
Connector Selection Decision Tree
Follow this tree to select the correct connector for any workflow.
Step 1: What is the source application?
Application
Connector
Direction
Skill
Revit
Revit Connector
Publish + Load
speckle-impl-revit
Rhino
Rhino Connector
Publish + Load
speckle-impl-rhino-grasshopper
Grasshopper
GH Connector (bundled with Rhino)
Publish + Load
speckle-impl-rhino-grasshopper
Blender
Blender Connector
Publish + Load
speckle-impl-blender
AutoCAD
AutoCAD Connector
Publish + Load
speckle-impl-autocad-civil3d
Civil 3D
Civil 3D Connector
Publish + Load
speckle-impl-autocad-civil3d
Tekla
Tekla Connector
Publish ONLY
speckle-impl-tekla
Archicad
Archicad Connector
Publish + Load
speckle-impl-connectors-overview
SketchUp
SketchUp Connector
Publish + Load
speckle-impl-connectors-overview
Power BI
Power BI Connector
Load ONLY (read)
speckle-impl-powerbi
Python script
SpecklePy
Publish + Load
speckle-impl-python-sdk
.NET application
Speckle.Sdk
Publish + Load
speckle-impl-sharp-sdk
Web application
@speckle/viewer
View only
speckle-impl-viewer
Step 2: What is the data flow direction?
Is data going INTO Speckle?
YES --> Use "Publish" (ToSpeckle conversion)
Consult the source connector skill
NO --> Is data coming OUT of Speckle?
YES --> Use "Load" (ToHost conversion)
Consult the target connector skill
CHECK: Does the target connector support Load?
Tekla: NO (publish-only)
Power BI: read-only (no write-back)
NO --> Is this viewing/analysis only?
YES --> Use Viewer or Power BI
NO --> Use GraphQL API for metadata operations
Step 3: Check fidelity constraints
Before confirming a workflow, ALWAYS check:
Native reconstruction: Loading into Revit ALWAYS creates Direct Shapes. Loading into Archicad ALWAYS creates GDL Objects. NEVER promise native element recreation.
Property survival: Custom properties are DROPPED on load by Revit, Blender, and Archicad. Properties persist in Speckle but NOT in the host application.
Geometry conversion: AutoCAD/Civil 3D solids become Mesh irreversibly. Brep geometry only survives in the Rhino ecosystem.
Texture loss: Textures NEVER transfer. Only RenderMaterial properties (color, opacity, metallic, roughness) survive.
Workflow Planning Templates
Template 1: Revit to Grasshopper Analysis
Use case: Extract Revit geometry for parametric analysis in Grasshopper, optionally push results back.
Step 1: Publish from Revit
Connector: Revit
Skill: speckle-impl-revit
Settings: Set reference point (Internal Origin recommended)
Output: RevitObjects with displayValue, properties, applicationIds
Step 2: Load in Grasshopper
Connector: Grasshopper (bundled with Rhino)
Skill: speckle-impl-rhino-grasshopper
Input: Speckle Receive component → project URL or model URL
Output: Data Objects (geometry + properties accessible via Deconstruct)
Step 3: Analyze/Modify in Grasshopper
ALWAYS use Passthrough nodes to preserve applicationId
NEVER create new objects from scratch (breaks change tracking)
Step 4 (optional): Publish results back
Connector: Grasshopper Send component
Output: Modified objects retain applicationId for version tracking
Step 5 (optional): Load back in Revit
WARNING: All objects become Direct Shapes
Native Revit properties are LOST as editable parameters
Error recovery: If Grasshopper loses applicationId, check that Passthrough nodes are used. If objects appear at wrong position in Revit, verify reference point settings match between publish and load.
Template 2: Multi-Discipline Coordination
Use case: Multiple teams working in different tools, need a shared federated view.
Step 1: Establish project structure
Skill: speckle-impl-versioning
Create ONE project with separate models per discipline:
architecture/ (Revit - architect)
structure/ (Tekla - structural engineer)
mep/ (Revit - MEP engineer)
landscape/ (Rhino - landscape)
Step 2: Align coordinate systems BEFORE publishing
Skill: speckle-impl-federation
CRITICAL: All Revit files MUST use the same reference point setting
Rhino/AutoCAD: Use World Origin (default)
Revit: Choose ONE setting (Survey Point recommended for geo-referenced)
Step 3: Each discipline publishes to their model
Each team uses their respective connector
ALWAYS include descriptive version messages
ALWAYS set sourceApplication for audit trail
Step 4: Federated view
Option A: Speckle Web Viewer (load multiple models)
Option B: Power BI with Speckle.Models.Federate() function
Option C: Custom viewer using @speckle/viewer
Step 5: Coordination review
Use viewer filtering to isolate disciplines
Use version diff to track changes between reviews
Error recovery: If models appear misaligned, check reference point settings per connector. If a discipline cannot see another's data, verify project permissions.
Template 3: Design Option Comparison
Use case: Compare multiple design iterations or alternatives.
Step 1: Choose model organization
Option A: Single model, multiple versions (same tool, sequential iterations)
Option B: Multiple models (different tools or parallel alternatives)
Step 2: Publish each option
Each publish creates an immutable version
ALWAYS use descriptive version messages:
"[DISCIPLINE]-[PHASE]-[DATE]-[DESCRIPTION]"
Example: "ARCH-SD-20260320-courtyard-option-a"
Step 3: Compare in viewer
Skill: speckle-impl-viewer (DiffExtension)
Skill: speckle-impl-versioning (version comparison)
Load two versions side-by-side or use diff overlay
Step 4: Document decision
Version history preserves all options permanently
Previous versions are NEVER overwritten
Template 4: Automated Quality Check
Use case: Run automated checks on every new version.
Step 1: Create Automate function
Skill: speckle-syntax-automate
Skill: speckle-impl-automate-functions
Write a function that validates incoming data
Step 2: Configure trigger
Trigger: "version created" on target model
The function runs automatically on each new version
Step 3: Access version data
Use automation_context to get the triggering version
Traverse objects, check properties, validate geometry
Step 4: Report results
Attach results to the version as Automate run output
Results visible in the web UI alongside the version
Version Strategy Decision Matrix
Scenario
Model Strategy
Version Naming
Skill
Single-discipline, sequential
One model, many versions
[DATE]-[DESCRIPTION]
speckle-impl-versioning
Multi-discipline coordination
One model per discipline
[DISCIPLINE]-[PHASE]-[DATE]
speckle-impl-federation
Design alternatives (same tool)
One model, tagged versions
option-[LETTER]-[DATE]
speckle-impl-versioning
Design alternatives (different tools)
One model per alternative
[TOOL]-option-[LETTER]
speckle-impl-federation
Campus/multi-building
One model per zone
[ZONE]-[DISCIPLINE]
speckle-impl-federation
CI/CD automated pipeline
Dedicated automation model
auto-[TIMESTAMP]-[HASH]
speckle-impl-automate-functions
Cross-Tool Fidelity Quick Reference
What survives when data moves between tools:
Data Type
Survives?
Constraint
Mesh geometry
ALWAYS
Universal primitive for all connectors
Brep/NURBS geometry
Rhino ecosystem ONLY
All other connectors get Mesh fallback
Material properties
ALWAYS
Color, opacity, metallic, roughness via RenderMaterial proxy
Textures
NEVER
Not supported in any connector
Custom properties
VARIES
Revit, Blender, Archicad DROP on load
Level associations
BIM connectors ONLY
Via Level proxy
Block/Instance definitions
Most connectors
Via Definition proxy
Named views
3D perspective ONLY
No plans, sections, elevations
Native type info
NEVER on load
Direct Shapes / GDL Objects only
applicationId
ALWAYS preserved in Speckle
May break in Grasshopper without Passthrough nodes
Error Recovery Playbook
Workflow step fails: objects appear at wrong position
Diagnose: Reference point mismatch between source and target
Check: Revit reference point setting (Internal Origin / Project Base / Survey Point)
Fix: Republish with consistent reference point across all files
