Crosstech Agents Aec Orchestrator

crosstech-agents-aec-orchestrator

Quick Reference

Routing Summary

Critical Warnings

NEVER skip the data integrity validation between hops — silent data loss at boundaries is the #1 cause of pipeline failures.

NEVER chain more than 4 hops without intermediate validation checkpoints — error propagation makes debugging impossible.

NEVER assume a single skill covers a multi-technology request — ALWAYS decompose into individual boundary crossings first.

ALWAYS load crosstech-core-ifc-schema-bridge when ANY hop involves IFC data.

ALWAYS load crosstech-core-coordinate-systems when ANY hop involves georeferenced data or coordinate transformations.

ALWAYS load the relevant crosstech-errors-* skill when a hop fails — do NOT guess at the fix.

Skill Routing Table

Wave A: Available Skills (9 skills)

Wave B: Deferred Skills (6 skills)

ALWAYS inform the user when a required skill is deferred and state which package must be installed.

Decision Tree: Single-Hop vs Multi-Hop

START: User request involves AEC technologies
│
├── Q1: How many technology boundaries are crossed?
│   ├── ONE boundary → Single-hop. Load the matching skill from the routing table.
│   └── TWO or more boundaries → Multi-hop. Continue to Q2.
│
├── Q2: Decompose into individual boundary crossings.
│   ├── List each hop: A → B → C → ...
│   ├── For EACH hop, identify the matching boundary skill.
│   └── Are ALL required skills available (Wave A)?
│       ├── YES → Proceed to Multi-Hop Pipeline Patterns.
│       └── NO → Identify which skills are DEFERRED (Wave B).
│           ├── Inform the user which packages must be installed.
│           └── Provide the available hops and mark gaps.
│
├── Q3: Does ANY hop involve IFC data?
│   ├── YES → ALWAYS also load crosstech-core-ifc-schema-bridge.
│   └── NO → Skip.
│
├── Q4: Does ANY hop involve coordinate transformations?
│   ├── YES → ALWAYS also load crosstech-core-coordinate-systems.
│   └── NO → Skip.
│
└── Q5: Is this a new pipeline design or debugging an existing one?
    ├── NEW DESIGN → Follow Multi-Hop Pipeline Patterns below.
    └── DEBUGGING → Load the relevant crosstech-errors-* skill first.

Multi-Hop Pipeline Patterns

Pattern 1: IFC → ERPNext Costing Pipeline

Scenario: Extract quantities from an IFC model and generate cost estimates in ERPNext.

IFC File
  │
  ├── Hop 1: Parse IFC structure
  │   Skill: crosstech-core-ifc-schema-bridge
  │   Action: Identify IfcQuantityArea, IfcQuantityVolume, IfcQuantityLength
  │   Validate: Schema version matches tool support, quantities exist
  │
  └── Hop 2: Map quantities to ERPNext
      Skill: crosstech-impl-ifc-erpnext-costing
      Action: Map property sets to BOM items, create cost estimate
      Validate: All mapped fields have values, unit conversions applied

Integrity checks between hops:

1. Verify IFC schema version is supported (IFC4 or IFC4X3)
2. Confirm quantity sets exist on target elements (IfcElementQuantity attached)
3. Verify unit consistency — IFC units MUST be converted before ERPNext mapping

Pattern 2: FreeCAD → IFC → Docker Processing

Scenario: Author a BIM model in FreeCAD, export to IFC, process in a Docker container.

FreeCAD Model
  │
  ├── Hop 1: Export IFC from FreeCAD
  │   Skill: crosstech-impl-freecad-ifc-bridge
  │   Action: Use NativeIFC mode, export to IFC4
  │   Validate: Exported IFC opens in IfcOpenShell without errors
  │
  ├── Hop 2: Containerize processing
  │   Skill: crosstech-impl-docker-aec-stack
  │   Action: Run IfcOpenShell validation in Docker container
  │   Validate: Container exits with code 0, output IFC is valid
  │
  └── Hop 3 (optional): Automate with n8n
      Skill: crosstech-impl-n8n-aec-pipeline
      Action: Trigger Docker processing on file upload
      Validate: n8n workflow completes, output file accessible

Integrity checks between hops:

1. After FreeCAD export: verify model.schema returns expected version
2. After Docker processing: verify file size is non-zero, schema unchanged
3. After n8n trigger: verify HTTP status 200, output path exists

Pattern 3: Full AEC Pipeline (n8n Orchestrated)

Scenario: Automated pipeline — IFC uploaded, validated, quantities extracted, costs pushed to ERPNext.

IFC Upload (webhook)
  │
  ├── Hop 1: n8n receives webhook trigger
  │   Skill: crosstech-impl-n8n-aec-pipeline
  │   Action: Webhook node receives file, stores to volume
  │   Validate: File exists on disk, size > 0
  │
  ├── Hop 2: Docker validates IFC
  │   Skill: crosstech-impl-docker-aec-stack
  │   Action: IfcOpenShell container validates schema + geometry
  │   Validate: Validation report has zero critical errors
  │
  ├── Hop 3: Parse IFC structure
  │   Skill: crosstech-core-ifc-schema-bridge
  │   Action: Extract entity hierarchy, property sets, quantities
  │   Validate: Expected entity types found, quantities non-empty
  │
  └── Hop 4: Push costs to ERPNext
      Skill: crosstech-impl-ifc-erpnext-costing
      Action: Map quantities → BOM items → cost estimate
      Validate: ERPNext API returns 200, estimate ID received

Integrity checks between hops:

1. After upload: file hash matches sender hash (no corruption)
2. After validation: zero CRITICAL findings, WARNINGS logged
3. After parsing: element count > 0, at least one quantity set found
4. After cost push: ERPNext response contains valid document name

Data Integrity Validation

Between-Hop Validation Checklist

ALWAYS run these checks after EACH boundary crossing:

Validation Code Pattern

def validate_hop(input_model, output_model, hop_name: str) -> list[str]:
    """ALWAYS call between pipeline hops. Returns list of errors (empty = pass)."""
    errors = []

    # Schema version
    if input_model.schema != output_model.schema:
        errors.append(f"[{hop_name}] Schema changed: {input_model.schema} → {output_model.schema}")

    # Entity count
    input_count = len(input_model.by_type("IfcProduct"))
    output_count = len(output_model.by_type("IfcProduct"))
    if output_count < input_count:
        errors.append(f"[{hop_name}] Entity loss: {input_count} → {output_count} ({input_count - output_count} lost)")

    # Property spot check (3 random elements)
    import random
    products = output_model.by_type("IfcProduct")
    if products:
        samples = random.sample(products, min(3, len(products)))
        for elem in samples:
            psets = ifcopenshell.util.element.get_psets(elem)
            if not psets:
                errors.append(f"[{hop_name}] {elem.is_a()} #{elem.id()} has no properties")

    return errors

Error Recovery

When a hop fails, follow this procedure:

HOP FAILED
│
├── Step 1: Identify the failure type
│   ├── Conversion error (missing geometry, lost properties, schema mismatch)
│   │   └── Load: crosstech-errors-conversion
│   ├── Coordinate error (wrong location, scale, axis, CRS)
│   │   └── Load: crosstech-errors-coordinate-mismatch
│   └── Tool-specific error (API error, timeout, auth failure)
│       └── Load the impl/ skill for that specific boundary
│
├── Step 2: Diagnose using the loaded error skill
│   └── Follow the symptom → root cause decision tree in that skill
│
├── Step 3: Fix and retry the failed hop ONLY
│   └── NEVER restart the entire pipeline — resume from the failed hop
│
└── Step 4: Re-run validation on the fixed hop output
    └── ONLY proceed to the next hop after validation passes

NEVER retry a failed hop more than 3 times without human review.

ALWAYS log the error details before retrying — silent retries mask root causes.

Critical Rules

1. ALWAYS decompose multi-technology requests into individual boundary crossings before selecting skills.
2. ALWAYS load crosstech-core-ifc-schema-bridge when ANY hop in the pipeline involves IFC data.
3. ALWAYS load crosstech-core-coordinate-systems when ANY hop involves georeferenced data.
4. ALWAYS validate data integrity between every hop using the validation checklist.
5. ALWAYS inform the user when a required skill is deferred (Wave B) and name the missing package.
6. ALWAYS load the relevant crosstech-errors-* skill when a hop fails — do NOT guess at fixes.
7. NEVER chain more than 4 hops without intermediate validation checkpoints.
8. NEVER skip between-hop validation — silent data loss at boundaries is the #1 pipeline failure cause.
9. NEVER retry a failed hop more than 3 times without escalating to human review.
10. NEVER assume two technologies use the same units, axis conventions, or schema versions.

Deferred Skills (Wave B)

The following 6 skills are planned but NOT YET AVAILABLE. They require their respective skill packages to be installed first.

Reference Links

• references/methods.md — Skill routing table details, validation method signatures
• references/examples.md — Complete multi-hop pipeline examples with code
• references/anti-patterns.md — Orchestration mistakes and how to avoid them

Related Skills (Cross-References)

• crosstech-core-ifc-schema-bridge — Foundation for ALL IFC-related hops
• crosstech-core-coordinate-systems — Foundation for ALL georeferenced hops
• crosstech-errors-conversion — First response for conversion failures
• crosstech-errors-coordinate-mismatch — First response for coordinate failures