earthtojake/text-to-cad
11 skills13.3k installs63.9k starsGitHub
Install
npx skills add https://github.com/earthtojake/text-to-cadSkills in this repo
1Cadcad is an agent skill for solo builders and makers who need machine-readable CAD from conversational requirements instead of clicking through parametric tools alone. It instructs the agent to capture a structured CAD brief—model intent, units, coordinate convention, overall dimensions, functional features, manufacturing assumptions, and mating or joint requirements—then produce and validate explicit CAD files and selector references rather than ambiguous geometry prose. The workflow emphasizes inspectable artifacts with defined STEP source and target paths, validation targets, and optional links for CAD Viewer review when the companion cad-viewer skill is installed. Invoke it when you are validating a hardware idea, preparing files for CAM or printing, or iterating dimensions before committing to expensive runs. It is generative and verification-oriented: outputs should be checkable against stated tolerances and positioning requirements. Pair it with cad-viewer for visual confirmation. The skill does not replace professional mechanical engineering sign-off for safety-critical parts, but it compresses the loop from text brief to reviewable model for indie hardware experiments and ag2.1kinstalls2Step PartsSTEP Parts is an agent skill that connects your coding assistant to the step.parts catalog API so you can locate and download STEP exchange files for CAD workflows. Solo and indie builders working on hardware prototypes, OpenSCAD-adjacent pipelines, or text-to-CAD experiments install it when they need real part geometry instead of placeholder meshes. Use it during build when you already know the part family or SKU area and want the agent to query step.parts and pull a file into the repo or build folder. It is narrower than a full MCAD suite: it does not model parts, only finds and downloads existing STEP assets from the documented API surface. Treat API origins and licensing on downloaded models as your responsibility; the skill documents defaults but does not validate fit or tolerances.1.8kinstalls3SdfSDF is an agent skill for solo builders working on robotics, simulation, or CAD-adjacent pipelines who need repeatable SDFormat models instead of hand-edited XML. It drives generation and updates from Python, runs validation on the resulting SDF, and can route validated models to a CAD viewer companion skill. Builder helpers are optional guardrails for XML structure—they supplement SDFormat and libsdformat rather than hiding the format. Use it when your agent stack already includes text-to-cad or Gazebo-style workflows and you want procedural model creation with a clear validate-then-preview path. Complexity sits at intermediate because you should understand links, inertial frames, and simulator expectations even when the agent drafts the files.1.7kinstalls4Urdfurdf is an agent skill for solo builders and robotics tinkerers who describe mechanisms in code and need consistent Unified Robot Description Format output without silent frame or axis errors. It emphasizes reading the URDF Design Ledger first so joints, links, and transforms stay coherent, then generating or updating URDF from Python sources and validating descriptions before commit. The packaged interface expects agents to invoke $urdf for generation and validation and to pass generated URDFs to $cad-viewer when that skill is installed, keeping text-to-CAD workflows connected end to end. It suits prototype arms, mobile bases, and simulation-ready assets during Build when hardware or sim integration is the goal rather than generic app UI work.1.7kinstalls5SrdfSRDF is an agent skill for solo and indie builders who are shipping robot motion stacks with MoveIt2 and need Semantic Robot Description Format files that planners can actually trust. It focuses on generating and validating disabled-collision and related semantics so SRDF is treated as derived evidence from your kinematic model, not as copy-pasted XML. When your agent is calibrating a manipulator or mobile arm pipeline, invoke it after you have link geometry and before you rely on collision-free planning in simulation or on hardware. The skill’s interface explicitly routes finished SRDF artifacts to cad-viewer for human review and optional MoveIt2 control tuning, which keeps the loop visual and debuggable for one-person teams. Expect intermediate familiarity with ROS2, MoveIt2, and SRDF concepts; the payoff is fewer mystery collisions and faster iteration between CAD sources and planning configuration.1.6kinstalls6Sendcutsendsendcutsend is a specialized agent skill for makers and hardware-minded solo builders who order laser-cut or CNC parts through SendCutSend. Instead of guessing whether a DXF or STEP will clear upload validation, you invoke the skill to walk the file against SendCutSend’s published spec sources, with explicit instruction to treat those sources as live evidence that can change field coverage and value types. The workflow emphasizes evidence-based review: cite the exact source URL (and JSON context where relevant) before stating whether a dimension, material rule, or geometry constraint passes or fails. It fits the validate prototype moment—after CAD export, before checkout—when a single spec mismatch can waste days and material cost. Pair it with your text-to-cad or manual export pipeline; it is a checker, not a geometry editor.1.6kinstalls7Renderrender is an agent skill for the text-to-cad ecosystem that starts or reuses CAD Explorer so you can visually inspect mechanical and robotics files on localhost. It expects concrete paths to files that already exist—typically right after another skill emitted STEP, STL, GLB, URDF, SRDF, or SDF output. The workflow centers on `dev:ensure` as the liveness and port-reuse gate; agents must not spin raw Vite dev servers that fork duplicate Explorer instances. Supported formats span exchange CAD, meshes, DXF, and robot descriptions. Solo hardware and robotics builders use it to close the loop between generation and human review, with snapshot links suitable for async feedback before fabrication or simulation handoff.1.4kinstalls8Cad Viewercad-viewer is an agent skill from the text-to-cad ecosystem that launches CAD Viewer so solo builders can sanity-check mechanical and robotics artifacts before sending them to print, simulation, or deployment. The primary operator flow is narrow: when the user supplies concrete CAD, G-code, or robot description files, the agent starts the viewer and hands back review links rather than improvising previews in chat. A second mode documented in the readme is strictly for contributors who want to change CAD Viewer itself—source work happens in the root development workspace and should not run during routine file review. Hardware-minded indies pairing LLM-assisted CAD with physical output use this skill to close the loop between generated geometry and human visual verification. It fits agents that can spawn local apps and resolve file paths on the developer machine.525installs9Gcodegcode is an agent skill for makers and indie builders who prototype physical parts, not typical web SaaS. It connects mesh assets to printer-ready G-code using a default prompt pattern ($gcode) and documents what automated validation can and cannot guarantee. The validation path is explicitly static: it catches structural problems defined in the skill but does not model acceleration limits, live firmware, or full slicer physics—so you still dry-run risky jobs on hardware you trust. Use it when you are closing the loop from CAD or scan meshes to a file your printer profile expects. It fits Prism’s audience when hardware, enclosures, or jigs are part of the product story alongside software.455installs10Bambu LabsBambu Labs is an agent skill for solo builders who ship physical parts from CAD or text-to-CAD pipelines and need a disciplined last mile to a Bambu printer on the LAN. Instead of blindly pushing G-code, the skill orients your agent around a dry-run handoff followed by a controlled start, which matches how you de-risk motion, temperatures, and file integrity before a long print. The packaged interface describes dry-run and start paths for validated plain G-code, while the bundled protocol notes stress that Bambu does not expose a stable public local-print contract—so you should plan and debug against your actual printer rather than treating LAN commands as guaranteed. That makes the skill best for indie hardware iteration, jigs, enclosures, and prototype runs where one bad start wastes hours and material. It is not a slicer replacement and does not remove the need for sound G-code upstream; it coordinates how your coding agent hands off to local Bambu workflows safely.436installs11Implicit CadImplicit CAD is an agent skill for solo builders who want lightweight 3D form exploration without OpenSCAD installs or heavy mesh editors. It steers agents to write browser-rendered implicit CAD as `.implicit.js` files built from GLSL signed-distance field primitives, then capture snapshots for sharing in issues, landing pages, or print prep discussions. The interface metadata sets display name, short description, and a default prompt so Claude Code, Cursor, or Codex consistently invoke the same capability. A Node entrypoint pattern supports local script execution when you need to batch or validate files on disk. Use it when you are validating enclosure shapes, abstract product visuals, or printable concepts before investing in STEP exports or vendor quotes. Complexity is advanced because SDF composition and shader debugging reward graphics comfort; beginners can still ship simple primitives by following agent-generated scaffolding.3installs