---
name: simplecadapi
description: Thin SimpleCAD SDK reference skill focused on the public API surface, core types, and current modeling workflows.
license: AGPL-3.0
compatibility: Documentation/reference bundle for current SimpleCADAPI surfaces.
metadata:
  project: simplecadapi
  version: 2.0.2
  package-name: simplecadapi
  package-version: 2.0.2
---

# SimpleCAD SDK Skill

## Philosophy
- This is a thin SDK reference skill: docs only.
- SDK source code is not bundled in this skill.

## Working From Repo Root
- Tool calls run from the repo root.
- Use one explicit skill root: `./skills/simplecadapi/` or `./workspace/skills/simplecadapi/`.
- Main doc paths:
  - `<skill_root>/SKILL.md`
  - `<skill_root>/references/docs/api/README.md`
  - `<skill_root>/references/docs/api/<api_name>.md`
  - `<skill_root>/references/docs/stdlib/README.md`
  - `<skill_root>/references/docs/stdlib/<stdlib_api_name>.md`
  - `<skill_root>/references/docs/core/<type_name>.md`
  - `<skill_root>/references/SDK_OVERVIEW.md`
  - `<skill_root>/references/SDK_SURFACES.md`
  - `<skill_root>/references/MODELING_WORKFLOWS.md`
  - `<skill_root>/references/inverse_engineer/brep-reverse-engineering.md`

## MUST Requirements
1. Read `SKILL.md`, `references/docs/api/README.md`, and `references/docs/stdlib/README.md` before choosing APIs.
2. Read the exact API Markdown page for every API you use.
3. Read the needed `core/` or exact `api/` docs when an API needs `Edge`, `Face`, `Wire`, `Solid`, `GraphSession`, `Sketch`, or expression types.
4. Prefer the standard parts library for standard parts before hand-modeling with core geometry APIs.
5. Follow the documented API signatures exactly.
6. When calling any SimpleCAD public API or standard-library function, use keyword arguments for every documented parameter; do not use positional arguments.
7. Use one `@model` entry point for replayable tasks, `@requires_session` for child builders, `capture_result(...)` for explicit outputs, and the returned `ModelResult` for model/session JSON and replay.
8. Use geometry APIs for integrated parts: profiles, features, booleans, transforms, tagging, QL inspection, serialization, and exports.
9. Use tags through `apply_tag(shape=..., tag=...)`, `apply_tag_rselection(scope=..., targets=..., tag=...)`, `list_tags(shape=..., scope=...)`, and `explain_tag(shape=..., tag=..., scope=...)`; do not call shape member tag mutators.
10. Build and validate incrementally. Each step MUST include a small grounding `print`, and grounding MUST use QL where possible.
11. For inspection/debugging, query geometry with QL and print only the queried facts you need; do not print whole solids or full model objects.
12. Boolean operations return a single `Solid`.
13. Use `union_rsolid(...)` for boolean union.
14. For automated example/test harnesses, prefer the repo-local examples in `examples/` and avoid scratch scripts in `sandbox/`.
15. If union cannot produce exactly one merged solid, it fails explicitly; do not silently pick one piece.
16. If a single merged solid is required and union fails, slightly adjust part placement so intended bodies overlap/embed, then recompute.
17. If a task depends on model replay or interchange, prefer `ModelResult.model_json` or `export_model_json()` output over hand-written payloads.
18. For STEP/BREP reverse engineering, geometry reconstruction, topology matching, feature-history inference, or target/candidate CAD comparison, read `references/inverse_engineer/brep-reverse-engineering.md` completely before inspecting or modeling.
19. In BREP reverse engineering, geometry equality and topology equality are hard requirements. Use engineering prior only to choose among processes that satisfy both; do not accept visual, volume, area, or topology-count similarity as completion.

## Task Router
- **Standard mechanical part**: read `references/docs/stdlib/README.md`, then the exact stdlib API page.
- **General integrated modeling**: read `references/MODELING_WORKFLOWS.md` and each exact API page used.
- **STEP/BREP reverse engineering or exact target comparison**: read `references/inverse_engineer/brep-reverse-engineering.md`; use `simplecadapi.inverse_engineer.brep` for inspection, strict comparison, views, and slice XOR diagnostics.
- **Sketch and constraints**: read the relevant sketch API pages and `references/docs/core/declarative_constraints.md` when needed.
- **Model JSON, graph replay, or translation**: read the graph/serialization API pages and the relevant translator documentation.

## Standard Parts Library
- SimpleCAD includes a standard library for parameterized mechanical parts.
- When the user needs a standard part and does not require complex custom geometry changes, use a standard-library function first.
- Current package-level standard-library surfaces include `scad.std.gear` for gears and racks, `scad.std.bearing` for ball bearing assemblies, `scad.std.chain` for roller-chain sprockets, and `scad.std.fastener` for bolts and nuts.
- Read `references/docs/stdlib/README.md` to discover standard-library functions.
- Read `references/docs/stdlib/<function_name>.md` before calling a standard-library function.
- Standard-library functions return normal SimpleCAD shapes or product assemblies that can be transformed, tagged, assembled, exported, and used with graph/model JSON workflows.

## Boolean result discipline
- `union_rsolid(...)`, `cut_rsolid(...)`, and `intersect_rsolid(...)` accept mixed inputs: standalone `Solid`, lists of `Solid`, and nested sequences.
- They return a single `Solid`.
- `union_rsolid(...)` already applies the package's default glue mode and a conservative internal tolerance.
- If a union cannot produce exactly one merged solid, it fails explicitly instead of returning multiple pieces.
- If a single merged solid is required but union fails, slightly move the parts so they overlap instead of merely touching, then recompute the union.

## Modeling Mental Model
- Start with intent: identify the part, its reference axes, critical profiles, and the features that produce the final solid.
- Build from lower-dimensional geometry to higher-dimensional geometry: `Vertex` / `Edge` / `Wire` / `Face` profiles first, then `Solid` features such as extrude, revolve, loft, and sweep.
- Keep modeling operations functional. Create new values from public functions such as `make_circle_rface(...)`, `extrude_rsolid(...)`, `cut_rsolid(...)`, and `fillet_rsolid(...)`.
- Use keyword arguments for all SimpleCAD function calls, for example `make_box_rsolid(width=10.0, height=20.0, depth=3.0)` instead of positional arguments.
- Use `@model` when the top-level model should be replayable, inspectable, exported as model JSON, or translated to another CAD system. It owns one `GraphSession`; reusable graph-producing builders use `@requires_session`.
- Treat model JSON as the interchange boundary. Prefer `ModelResult.model_json` and `ModelResult.replay()` for top-level models; use `export_model_json(session=...)` for lower-level direct sessions and `replay_model_json(json_str=...)` for standalone payloads.
- Use QL for precise grounding. Query faces, edges, centers, normals, areas, lengths, curve types, and tags; print only the facts needed to validate the current step.
- Use `get_edges(index)`, `get_faces(index)`, `get_wires(index)`, or `get_vertices(index)` when an indexed topology pick is intentional; these picks are preserved as geo select nodes in replayable graph workflows.
- Use tags for topology identity, semantic intent, and selection anchors, such as `housing.face.top`, `role.mounting_surface`, `anchor.datum.primary`, or `group.fasteners`.
- Keep numeric and geometric facts in metadata or graph payloads, not in tags.
- When a QL-selected face or edge is used by a later feature, expect the graph/model workflow to preserve that selection as a stable geo select node.
- For FreeCAD translation, prefer canonical model JSON generated from a `GraphSession`; selected profiles and detail-feature selections should come from the graph rather than ad hoc object lookup.

## Tagging Mental Model
- Public tag attachment is `apply_tag(shape=..., tag=...)`.
- Multi-entity or explicitly propagated attachment is `apply_tag_rselection(scope=..., targets=..., tag=..., topology_propagation=..., lineage_policy=...)`; it returns an independent semantic shape view.
- Public tag inspection is `list_tags(shape=..., scope=...)`, which returns a stable sorted list. Use `explain_tag(...)` when producer and evidence matter.
- Tags are normalized lowercase dot-separated semantic tokens, for example `role.mounting_surface`, `anchor.datum.primary`, `group.fasteners`, `face.top`, or `solid.boolean.cut`.
- Do not encode numeric dimensions or descriptive geometry payloads in tags; store them in metadata such as `shape.get_metadata("geo")` or `shape.set_metadata(...)`.
- Direct user assignments default to local topology propagation regardless of tag text. Constructor-generated topology-identity Face tags may use downward propagation so boundary Edges expose the Face tag; use `TopologyPropagation.DOWNWARD` explicitly for the same behavior in `apply_tag_rselection(...)`.
- `effective` includes local and inherited bindings but excludes lineage. Use `scope=TagScope.LINEAGE` only when complete topology-history evidence is available.
- Operation events, source roles, and feature output roles are typed `metadata["track"]`, not flat tags. Query them with `ql.operation_event(...)`, `ql.origin_role(...)`, and `ql.output_role(...)`; unknown correspondence remains partial/unknown.
- `extrude_rsolid`, `revolve_rsolid`, `fillet_rsolid`, `chamfer_rsolid`, `shell_rsolid`, `loft_rsolid`, `sweep_rsolid`, and `twisted_sweep_rsolid` accept strict role-specific tag arguments. Requested roles must have complete kernel evidence and satisfy their documented cardinality or the whole call fails.
- Role-specific tag arguments and `result_tag` lower to canonical `apply_tag_rselection` nodes in a `GraphSession`; they are semantic assignments, not geometry parameters. There is one public argument per target role; do not use a generic role-to-tag mapping.
- Topology identity and user semantics use the same `TagBinding`, `list_tags(...)`, `explain_tag(...)`, and `ql.tag(...)` surfaces. One topology object may carry multiple tags for different purposes.
- Geometry constructors and features use `tag_prefix` to create topology-identity tags such as `housing.face.top`; profile APIs use `edge_tag` or `edge_tags` for local Edge tag segments. These bindings carry `topology_name` evidence and project only across exact kernel-proven correspondence.
- Role-specific `*_tag` arguments and `result_tag` create tags with operation-role or result evidence. Tag text alone does not determine evidence or projection policy.
- Query proven source projection with `ql.source_binding(...)` or `ql.source_topology(...)`. These predicates inspect canonical local binding evidence, never tag text or geometric similarity.
- When a tagged profile entity must feed a later feature, pass the semantic view returned by `apply_tag_rselection(...)` into that feature. Tagging a detached branch and then using the original profile does not create hidden graph coupling.
- Prefer scoped QL tag predicates (`ql.tag("role.*", scope="effective")`, `ql.select(...).where(...)`) for inspection and grounding.

## SDK Focus
- This skill is intended to describe the public CAD Python SDK surface.
- Prefer the generated API, stdlib, and core docs over environment/bootstrap instructions.
- API docs include an `Import Surface` section that distinguishes top-level exports, submodule APIs, and translator backend APIs under `simplecadapi.translator.<backend>`.
- Stdlib docs include an `Import Surface` section that identifies the package-level `simplecadapi.std.gear` module export.
- Use `references/SDK_OVERVIEW.md` for the package-level map.
- Use `references/SDK_SURFACES.md` for the main public surfaces.
- Use `references/MODELING_WORKFLOWS.md` for graph/model-oriented patterns.

## Example SDK usage

```python
import simplecadapi as scad
from simplecadapi import ModelResult, capture_result, model, requires_session
```

Typical replayable usage in a Python script:

```python
import simplecadapi as scad

@scad.model(graph_id="box")
def build_box():
    shape = scad.make_box_rsolid(width=10.0, height=20.0, depth=30.0)
    scad.capture_result(value=shape)
    return shape

result = build_box()
rebuilt = result.replay()
print(len(rebuilt))
```

Use the graph/model JSON workflow when the task needs reproducibility, interchange, or replayable outputs.

## References
- `references/SDK_OVERVIEW.md`
- `references/SDK_SURFACES.md`
- `references/MODELING_WORKFLOWS.md`
- `references/SDK_PACKAGE_SUMMARY.md`
- `references/docs/api/`
- `references/docs/stdlib/`
- `references/docs/core/`
- `references/inverse_engineer/brep-reverse-engineering.md`