cad-reference
Use when checking a CAD or printing anti-pattern, looking up a tolerance, temperature or dimensional value, asking what actually moved (OpenSCAD's Manifold backend, the slicer landscape, code-CAD and the LLM angle, verified August 2026), finding the books and tools, or needing a picker and a pre-print checklist. Companion to the other cad-3d-printing skills.
- License 许可
- MIT
- 最近更新
- 2026-11-15
- Stars
- ★ 5
- Forks
- 3
01 前言 Front Matter
- name
- cad-reference
- description
- Use when checking a CAD or printing anti-pattern, looking up a tolerance, temperature or dimensional value, asking what actually moved (OpenSCAD's Manifold backend, the slicer landscape, code-CAD and the LLM angle, verified August 2026), finding the books and tools, or needing a picker and a pre-print checklist. Companion to the other cad-3d-printing skills.
03 SKILL.md 正文
CAD and 3D Printing: Anti-Patterns, Numbers, What Moved, and Tools
Part 5 of 5 of the CAD and 3D Printing reference (plugin
cad-3d-printing), covering §13–§18. Sibling skills:cad-geometry-kernels-formats-and-code-cad(§0–§4),cad-slicing-pipeline-and-processes(§5–§6),cad-design-tolerances-and-materials(§7–§9),cad-generative-automation-and-scanning(§10–§12). Section numbers are shared across the set; a reference written as §N →skillpoints into that sibling skill.Currency: The geometry mathematics and process physics are stable; the code-CAD tooling and slicer landscape moved. See §16 below for both, dated.
Scope. Written for people who write code. §2–§4 →
cad-geometry-kernels-formats-and-code-cadare the software engineering (kernels, scripting, formats); §6–§9 →cad-slicing-pipeline-and-processes,cad-design-tolerances-and-materialsare the manufacturing physics you need for the software to produce parts that work.⚠️ GOTCHA boxes mark what silently produces a bad model or an unprintable part.
The three ideas that organize everything:
- ⚠️ Choose your representation before your tool. B-rep, mesh, and implicit fields are not interchangeable — each makes some operations trivial and others impossible, and most CAD frustration is a representation mismatch (§1 →
cad-geometry-kernels-formats-and-code-cad).- ⚠️ Every export loses something. STL loses units, topology, colour, and all parametric intent. STEP keeps the solid but not the feature tree. Know what falls off at each hop (§3 →
cad-geometry-kernels-formats-and-code-cad).- ⚠️ The printer does not build your model, it builds your G-code. Slicing is a lossy, opinionated transformation, and the part's real properties — strength, accuracy, surface — are determined there and by the process physics, not by the CAD (§5 →
cad-slicing-pipeline-and-processes, §6 →cad-slicing-pipeline-and-processes).
§13. Anti-Patterns
| Anti-pattern | Why |
|---|---|
| Choosing the tool before the representation | ⚠️ Most CAD frustration is a representation mismatch (§1 → cad-geometry-kernels-formats-and-code-cad) |
| Expecting arbitrary-edge fillets in OpenSCAD | ⚠️ CSG has no edges to select (§4.2 → cad-geometry-kernels-formats-and-code-cad) |
| Coplanar faces in a boolean | ⚠️ Numerically ambiguous. Overlap by an epsilon (§2 → cad-geometry-kernels-formats-and-code-cad) |
| STL as an interchange format between CAD tools | ⚠️ Lossy and unit-less. Use STEP (§3 → cad-geometry-kernels-formats-and-code-cad) |
| STL to the printer when 3MF is available | 3MF carries units, colour, settings (§3 → cad-geometry-kernels-formats-and-code-cad) |
| Assuming an STL's units | ⚠️ There are none. 25.4× errors are real (§3 → cad-geometry-kernels-formats-and-code-cad) |
| Round-tripping CAD → STL → CAD | You get an uneditable mesh back (§3 → cad-geometry-kernels-formats-and-code-cad) |
High $fn / fine tessellation while iterating |
⚠️ 2 seconds becomes 2 minutes (§4.3 → cad-geometry-kernels-formats-and-code-cad) |
| Magic numbers instead of named parameters | Defeats the point of code-CAD (§4.3 → cad-geometry-kernels-formats-and-code-cad) |
| Version-controlling the STL instead of the source | ⚠️ Same (§4.3 → cad-geometry-kernels-formats-and-code-cad) |
| Trusting automatic mesh repair blindly | ⚠️ Heuristic; silently changes geometry (§4.5 → cad-geometry-kernels-formats-and-code-cad) |
| Laplacian smoothing on a dimensioned part | ⚠️ It shrinks the model (§4.5 → cad-geometry-kernels-formats-and-code-cad) |
| Ignoring print orientation for structural parts | ⚠️ Z is 20–50% weaker. Orientation is a design decision (§6.1 → cad-slicing-pipeline-and-processes) |
| Horizontal circular holes on FDM | ⚠️ Sagging top. Use a teardrop (§7 → cad-design-tolerances-and-materials) |
| Wall thickness not a multiple of extrusion width | Gaps or over-extrusion (§7 → cad-design-tolerances-and-materials) |
| Fully enclosed voids in SLS or SLA | ⚠️ Trapped powder / resin, permanently (§6.2 → cad-slicing-pipeline-and-processes, §7 → cad-design-tolerances-and-materials) |
| Modelling holes at nominal size for FDM | ⚠️ They print undersize (§8 → cad-design-tolerances-and-materials) |
| Trusting a clearance table over a test print | Calibrate your machine and material (§8 → cad-design-tolerances-and-materials) |
| Printed threads for load-bearing joints | ⚠️ Use heat-set inserts (§8 → cad-design-tolerances-and-materials) |
| Printing hygroscopic filament without drying | ⚠️ The most common invisible cause of bad prints (§9 → cad-design-tolerances-and-materials) |
| Infill density as the strength lever | ⚠️ Perimeters do more above ~40% (§5.3 → cad-slicing-pipeline-and-processes) |
| Topology optimization on one load case | Fragile in every other direction (§10 → cad-generative-automation-and-scanning) |
| Manual slicing in a repeatable workflow | ⚠️ Every slicer has a CLI (§11 → cad-generative-automation-and-scanning) |
| Treating a scan as a model | ⚠️ A mesh is not CAD (§12 → cad-generative-automation-and-scanning) |
§14. Numbers
FDM
Nozzle 0.4 mm typical (0.2–0.8) · Layer 0.1–0.3 mm (⚠️ ~25–75% of nozzle Ø)
Extrusion width ≈ nozzle Ø to 1.2× · Filament 1.75 mm (or 2.85)
⚠️ Overhang limit ~45–55° · Bridging to ~50 mm with good cooling
⚠️ Z-strength 50–80% of XY · Accuracy ±0.1–0.5 mm
First layer squish, elephant's foot compensation ~0.1–0.5 mm
CLEARANCES (FDM, per side)
Press fit 0 to −0.1 · Tight sliding 0.15–0.2 · Free 0.3–0.4 · Print-in-place 0.3–0.5
⚠️ Holes print undersize: oversize by 0.1–0.4 mm
RESOLUTION BY PROCESS
Material jetting 16 µm · SLA/MSLA 25–100 µm · SLM 20–50 µm
MJF ~80 µm · SLS ~100 µm · FDM 100–300 µm
TEMPERATURES (nozzle / bed)
PLA 200/60 · PETG 240/80 · ABS 250/100 · Nylon 260/90 · PC 280/110 · TPU 225/50
⚠️ PLA softens ~60 °C — glass transition, not melting
EXTRUSION MATH
E = (layer_h × width × dist) / (π (filament_d/2)²)
⚠️ 1.75 mm nominal at 1.70 mm actual = ~6% under-extrusion
MESH
⚠️ Euler: V − E + F = 2 − 2g for a closed surface of genus g
Manifold: every edge in exactly 2 faces
§15. Books and Tools
| Source | Why |
|---|---|
| Piegl & Tiller, The NURBS Book | ⚠️ The reference for §1 → cad-geometry-kernels-formats-and-code-cad's surface math |
| Farin, Curves and Surfaces for CAGD | The readable companion |
| Botsch et al., Polygon Mesh Processing | ⚠️ §4.5 → cad-geometry-kernels-formats-and-code-cad, definitively |
| de Berg et al., Computational Geometry | Algorithms and the robustness problem |
| Shirley, Ray Tracing in One Weekend | ⚠️ The best free introduction to thinking in 3D geometry code |
| Gibson, Rosen & Stucker, Additive Manufacturing Technologies | ⚠️ The academic AM reference |
| Redwood et al., The 3D Printing Handbook | ⚠️ The practical DfAM book — get this one first |
| Bendsøe & Sigmund, Topology Optimization | §10 → cad-generative-automation-and-scanning |
| ASME Y14.5 (GD&T) | Tolerancing formally (§8 → cad-design-tolerances-and-materials) |
Tools: OpenSCAD, CadQuery, build123d, FreeCAD, SolveSpace (⚠️ an underrated constraint-based parametric modeller), Blender, Rhino + Grasshopper, nTop (implicits), Onshape (⚠️ best commercial API), Fusion 360. Mesh: trimesh, pymeshlab, MeshLab, Open3D, libigl, Manifold, OpenVDB. Slicers: §16.2. Print management: OctoPrint, Klipper/Moonraker, Mainsail, Fluidd. Communities: Printables, Thingiverse, MakerWorld, ⚠️ and the slicer projects' GitHub issues, which are the best documentation of real failure modes.
§16. What Actually Moved — verified August 2026
16.1 ⚠️ OpenSCAD's Manifold backend
The most significant change to open code-CAD in years, and one worth getting right because outdated descriptions are everywhere.
- OpenSCAD historically used CGAL, which is exact and ⚠️ notoriously slow.
- Manifold replaced it as an alternative backend. Reported speedups: 5–30× over CGAL's fast-csg, which was itself 30–150× over baseline CGAL Nef routines — ⚠️ with a 1,000× improvement reported on at least one model.
- ⚠️ Since 2024-09-28 the Manifold backend is no longer experimental — selectable in
Preferences → Advanced → 3D Rendering → Backend, or
openscad --backend=manifoldon the command line. ⚠️ The old feature-flag method was removed, which broke some command-line scripts. - ⚠️ At that announcement, CGAL was still the default backend, with Manifold opt-in.
- ⚠️ That changed on 2025-08-17, when OpenSCAD maintainer Marius Kintel announced on the
project mailing list that Manifold is now the default backend in development
snapshots, calling it "battle tested"; CGAL remains selectable via preferences or
--backend=cgal. ⚠️ The most recent tagged "stable" release is still 2021.01, which predates Manifold support entirely — anyone using Manifold at all is already on a development snapshot, where the default has now flipped.
⚠️ GOTCHA — two things to be careful about here. First, many current sources — including reference documentation and comparison articles — still describe OpenSCAD's kernel as simply "CGAL." That was accurate and is now at best incomplete. ⚠️ Check which backend you're actually running before attributing behaviour to a kernel. Second, the backends are not equivalent in output. A February 2026 bug report shows a model where Manifold produced STL with open edges that PrusaSlicer auto-repaired, while CGAL made most of the model vanish entirely — and ⚠️ the reporter's fix was adding a small overlap, i.e. the §2 →
cad-geometry-kernels-formats-and-code-cadepsilon problem. If a model behaves oddly, try the other backend as a diagnostic.
16.2 The slicer landscape
⚠️ Almost everything descends from Slic3r. OrcaSlicer is a fork of Bambu Studio, which forked PrusaSlicer, which began as Slic3r PE. Cura is the separate lineage.
As of 2026: OrcaSlicer is widely described as the strongest for advanced FDM tuning — ⚠️ best-in-class calibration workflows (temperature, flow, pressure advance, retraction, max volumetric speed), the broadest third-party printer support, native presets for Klipper input-shaping and pressure-advance, no cloud requirement, and leading tree/organic support generation. Bambu Studio is strongest on Bambu hardware, PrusaSlicer on Prusa hardware and mature multi-material with a 1–3 month-slower release cadence, and Cura remains the broad beginner-friendly standby with the plugin ecosystem.
⚠️ Read that as ecosystem fit, not a ranking: the honest summary from the comparisons is that the right slicer depends almost entirely on what you're printing on, and Klipper users in particular have a clear answer.
16.3 Code-CAD and the LLM angle
CadQuery and build123d wrap OCCT, giving exact B-rep, true fillets and chamfers, and STEP import/export that OpenSCAD cannot do. build123d replaces CadQuery's fluent method-chaining with context managers, so ordinary Python loops, references and filtering work naturally.
⚠️ One genuinely interesting current finding, reported by more than one party building LLM-to-CAD tooling: language models write OpenSCAD substantially more reliably than CadQuery or build123d — one benchmark reports 3–4× fewer code errors — attributed to training-data volume and to OCCT's complexity leaking through the Python wrapper.
⚠️ GOTCHA — take that with real caution. ⚠️ Both sources reporting it are companies whose product is built on OpenSCAD, so they are not disinterested. The underlying reasoning is plausible (OpenSCAD's language is small and its corpus is large), but this is a vendor-adjacent benchmark, not an independent one. ⚠️ And it says nothing about which kernel is better for a human — the same sources concede that OCCT is the more capable kernel.
§17. Quick Reference
17.1 Picker
| Need | Use |
|---|---|
| Parametric part, simple geometry, scriptable | OpenSCAD (§4.1 → cad-geometry-kernels-formats-and-code-cad) |
| Same, but needs fillets/chamfers and STEP | ⚠️ CadQuery or build123d (§4.2 → cad-geometry-kernels-formats-and-code-cad) |
| Full CAD app automation | FreeCAD Python, or Onshape REST (§4.4 → cad-geometry-kernels-formats-and-code-cad) |
| Constraint-based parametric, lightweight GUI | SolveSpace (§15) |
| Organic/artistic modelling | Blender (⚠️ not engineering CAD) (§4.1 → cad-geometry-kernels-formats-and-code-cad) |
| Lattices, implicit geometry, millions of features | ⚠️ Implicit/SDF representation (§1 → cad-geometry-kernels-formats-and-code-cad, §10 → cad-generative-automation-and-scanning) |
| Reliable booleans on messy meshes | Manifold (§2 → cad-geometry-kernels-formats-and-code-cad) |
| Repair a broken mesh | trimesh / pymeshlab / Netfabb; ⚠️ voxel remesh as last resort (§4.5 → cad-geometry-kernels-formats-and-code-cad) |
| CAD → CAD interchange | ⚠️ STEP (§3 → cad-geometry-kernels-formats-and-code-cad) |
| CAD → printer | ⚠️ 3MF (§3 → cad-geometry-kernels-formats-and-code-cad) |
| Batch/CI generation | ⚠️ Headless CLI — all of them have one (§11 → cad-generative-automation-and-scanning) |
| Klipper printer | OrcaSlicer (§16.2) |
| Strong part, FDM | ⚠️ Orient for load along layers; add perimeters, not infill (§5.3 → cad-slicing-pipeline-and-processes, §6.1 → cad-slicing-pipeline-and-processes) |
| Load-bearing threaded joint, FDM | Heat-set insert (§8 → cad-design-tolerances-and-materials) |
| Complex geometry, no supports | ⚠️ SLS/MJF — the powder bed supports it (§6 → cad-slicing-pipeline-and-processes) |
| Fine detail | SLA/MSLA (§6 → cad-slicing-pipeline-and-processes) |
17.2 Pre-print checklist
- [ ] Mesh manifold, watertight, correct normals? (§4.5 →
cad-geometry-kernels-formats-and-code-cad) - [ ] Units correct — and is it 3MF rather than STL? (§3 →
cad-geometry-kernels-formats-and-code-cad) - [ ] Oriented for the load path, not just for print time? (§6.1 →
cad-slicing-pipeline-and-processes) - [ ] Overhangs under ~45°, or supported, or redesigned as chamfers? (§7 →
cad-design-tolerances-and-materials) - [ ] Horizontal holes teardropped? (§7 →
cad-design-tolerances-and-materials) - [ ] Walls a whole multiple of extrusion width? (§7 →
cad-design-tolerances-and-materials) - [ ] Clearances calibrated for this machine and material? (§8 →
cad-design-tolerances-and-materials) - [ ] Escape holes for powder/resin if applicable? (§6.2 →
cad-slicing-pipeline-and-processes) - [ ] Filament dried if hygroscopic? (§9 →
cad-design-tolerances-and-materials) - [ ] Perimeters increased before infill for strength? (§5.3 →
cad-slicing-pipeline-and-processes)
§18. Method
§1–§12 → cad-geometry-kernels-formats-and-code-cad, cad-slicing-pipeline-and-processes, cad-design-tolerances-and-materials, cad-generative-automation-and-scanning rest on stable material — NURBS and B-rep mathematics, computational geometry,
mesh processing, and additive manufacturing process physics — sourced from the references
in §15, chiefly Piegl & Tiller, Botsch et al., and Gibson, Rosen & Stucker,
plus the file-format specifications. ⚠️ None of that needed web verification; the
geometry math is decades old and the process physics is thermodynamics.
Two searches were run in August 2026, confined to the tooling: the code-CAD ecosystem and the OpenSCAD kernel and slicer landscape.
⚠️ The second search materially corrected the first. Several sources returned by the code-CAD search — including CadQuery's own documentation and multiple 2026 comparison articles — describe OpenSCAD's kernel as CGAL and contrast it unfavourably with OCCT on that basis. That framing predates the Manifold work. I followed up specifically and found the OpenSCAD mailing list announcement (2024-09-28) that Manifold is no longer experimental, a later mailing list post (2025-08-17) from the same maintainer confirming Manifold became the default backend in development snapshots, plus the Manifold project's own performance discussion for the speedup figures. ⚠️ §16.1 reflects that; the widely-repeated "OpenSCAD uses CGAL" line is now incomplete.
Confidence. High in §1–§12 → cad-geometry-kernels-formats-and-code-cad, cad-slicing-pipeline-and-processes, cad-design-tolerances-and-materials, cad-generative-automation-and-scanning — established mathematics and process physics, with
numbers stated as representative ranges that vary by machine and material. High in
§16.1's Manifold facts, which come from the project and the OpenSCAD maintainers directly,
including that CGAL remained the default backend at the time of the 2024-09-28
announcement, and that Manifold became the default in development snapshots as of
2025-08-17 — ⚠️ verify which snapshot and backend you are actually running rather than
assuming, since the tagged "stable" release (2021.01) predates either backend option.
⚠️ Two explicit hedges. §16.2's slicer assessment is drawn from comparison articles, which are opinion-shaped and frequently affiliate-monetized; I have framed it as ecosystem fit rather than a ranking, and the technical lineage (Slic3r → PrusaSlicer → Bambu Studio → OrcaSlicer) is the part that is unambiguous. And ⚠️ §16.3's LLM benchmark is flagged in place as vendor-adjacent — both parties reporting it build products on OpenSCAD. The mechanism they propose is plausible and the finding may well be right, but it is not an independent evaluation and I have not treated it as one.
§14's tolerance and clearance figures are starting points, not specifications —
⚠️ the §8 → cad-design-tolerances-and-materials advice to run a calibration print is the actual recommendation, and it
supersedes any table including mine.
04 同仓库更多 Skills
white-paper-writing
Use when writing white papers, technical reports, policy documents, thought leadership pieces, or any long-form authoritative document. Triggers on requests to write, structure, outline, or review white papers, technical reports, position papers, policy briefs, or evidence-driven arguments. Also use for B2B content marketing strategy, technical documentation requiring authority, academic-style reports, or any document that must educate while persuading. Use when someone asks "how should I structure this" for a long-form document.
prose-technical-mechanics
Use when editing or writing prose at a technical/mechanical level: choosing between Anglo-Saxon and Latinate diction, placing information in topic vs stress positions, writing periodic vs cumulative sentences, handling sentence rhythm, punctuation as logic, transitions, paragraph structure, or phonesthetics. Use when someone wants to understand WHY prose works or doesn't, not just surface-level style advice. Triggers on word choice, sentence structure, emphasis, rhythm, diction, or any deep writing mechanics question.
prose-mechanics
Use when writing, editing, or analyzing prose style; when trying to avoid generic AI-sounding writing; when applying specific prose styles such as journalistic, academic, narrative, KJV biblical, Hemingway plain, Latinate/Jamesian, Orwellian, Augustan, Victorian, Early Modern (Bacon, Browne, Milton), or free indirect discourse; or when coaching someone on writing voice and mechanics. Triggers on any request to improve, analyze, or write in a specific prose style, to fix "AI-sounding" writing, to make prose more vivid or concrete, to vary sentence rhythm, or to understand how a particular writer achieves their effects.
narrative-structure
Use when structuring stories, pitches, presentations, white papers, or any persuasive narrative; analyzing the hero's journey in fiction or non-fiction; applying StoryBrand for business communication; designing narrative arcs for thought leadership; or understanding Vogler, Snyder, Propp, or Campbell frameworks. Triggers on any narrative structure, storytelling, pitch structure, StoryBrand, hero's journey, three-act structure, Freytag, Save the Cat, narrative arc, or "how should I structure this story/pitch/presentation" request.
legal-brief-writing
Use when writing legal briefs, motions, or any court filing; when analyzing judicial cognition and persuasion; when applying elite brief-writing techniques from Roberts, Clement, Kagan, Garner; or when structuring legal arguments for maximum persuasion. Also use for autistic/neurodivergent advocates who need to understand how their natural communication style maps to elite legal advocacy. Triggers on any legal writing, brief drafting, motion preparation, or legal argument structuring request.
journal-publishability-criteria
Use whenever a research paper, article or preprint is being written, revised, reviewed or published, and specifically before every publication of docs/paper.md. Covers what legitimate journals actually gate on, in order of gating — the editor's desk screen (scope, novelty, package problems), what reviewers judge (methodological soundness, claims the evidence can carry, statistical and qualitative reporting), EQUATOR reporting guidelines, the ICMJE authorship criteria, COPE's view of text recycling, the near-universal generative-AI policy (no AI author, disclosure, author responsibility), TOP 2025 transparency standards, how to verify a journal is legitimate, submitting without an institution, and a pre-submission checklist. Triggers on requests to evaluate, review, prepare or submit a paper, to judge publishability, to write a declarations or data-availability section, to disclose AI use, or to check a journal.