agentleFS
JiamingShi-JsPhysJiamingShi-JsPhys/3d-fabric/.github/copilot-instructions.md

3d-fabric

Copilot instructions2 starsChanged 3 days agoMITView on GitHub

3d-fabric is the whole 3D toolchain for this machine in one plugin, in two halves. The CAD half builds a part by natural language through SolidCraft, reads and edits a design someone already made, measures it, renders it, checks any mesh for printing, and repairs meshes that came from elsewhere. The Blender half ships thirty skills for organic modelling, materials, lighting, cameras, rendering, UV and animation. Pick the half by the job: a printable, dimensioned part is CAD; something with…

The file

We split this copilot-instructions.md into the rules it gives an agent

6 rules. About 0 are common, 1 is distinctive, and 5 are very specific to this project.

  • ≈0

    Common · estimate

    Many files like this have a rule of this kind.

  • ≈1

    Distinctive · estimate

    This author's own choices. Worth a look.

  • 5

    Very specific

    Tied to their project's names, files or tools. Probably skip.

  • Common
  • Distinctive
  • Very specific
  1. Very specific:

    Before the first Blender call, read skills/text-to-blender/references/mcp-backends.md.

    line 84
  2. Very specific:

    Never edit those files — they are upstream text.

    line 97
  3. Very specific:

    Do not invent SolidCraft command names.

    line 102
  4. Very specific:

    Treat error: step N (cmd) as: command N of the list (1-based), fix that one, re-run the whole script.

    line 110
  5. Very specific:

    Never modify the user's original .sc.

    line 112
  6. Distinctive:

    Never present a STEP export as printable: it is a B-rep with no triangles.

    line 114

≈ “Common” is an estimate: at least 1 in 5 of the 3,827 Copilot instructions files we read have a rule of the same kind, not the same words. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.

Keep what fits and drop the rest. Your version, ready for your agents.
Files
copilot-instructions.md

3d-fabric — agent operating rules (v1.1.2)

3d-fabric is the whole 3D toolchain for this machine in one plugin, in two halves. The CAD half builds a part by natural language through SolidCraft, reads and edits a design someone already made, measures it, renders it, checks any mesh for printing, and repairs meshes that came from elsewhere. The Blender half ships thirty skills for organic modelling, materials, lighting, cameras, rendering, UV and animation.

Pick the half by the job: a printable, dimensioned part is CAD; something with a silhouette, a surface look or a moving shot is Blender.

Install as one plugin

Install this repository as a single plugin. Do NOT copy run.py on its own — it imports scripts/engine.py and scripts/solidcraft.py, and the skill lives at skills/3d-fabric/SKILL.md. Use the marketplace, or let install.py --skill-dir perform the copy.

skills/ holds thirty-one skills: the CAD skill and the thirty Blender ones. Hosts that read a rule file instead of a plugin — Cursor, Windsurf, Cline, Continue, Roo Code, GitHub Copilot — get this same contract through python install.py --project <dir>; python install.py --list shows every destination. .claude-plugin/, .codebuddy-plugin/, .codex-plugin/ and .dsh-plugin/ are the four plugin manifests; this file is the body all six rule files carry.

Entry points (Python 3.10+)

python run.py --action probe                       # is SolidCraft reachable?
python run.py --action model   --input req.json    # build + export + inspect
python run.py --action script  --input part.sc --output build.json
python run.py --action model   --script patch.json --in part.sc --output new.stl
python run.py --action measure --input part.sc
python run.py --action snapshot --input part.sc --output preview.png --view iso
python run.py --action inspect --input part.stl    # mesh gate only
python run.py --action repair  --input part.stl --output fixed.stl

The seven actions

  1. probe — locate the CLI (--solidcraft, SOLIDCRAFT_CLI, SOLIDCRAFT_HOME, PATH, known folders) and verify it with --version. A missing CLI is UNVERIFIED, never a crash.
  2. model — run a {"commands": [...]} script or .sc design, export by extension, then inspect automatically. Keeps the executed script and an editable .sc beside the output. --in design.sc applies the script on top of an existing design; the original file is never modified.
  3. script — turn a .sc / .step back into {"commands": [...]}, so a design that arrived without its script can still be read, edited and replayed.
  4. measure — per-body volume, area, mass (from the design's material), bounds, closedness, face/edge/vertex counts, face types and all parameters. Answer "how heavy / how big" from this, not from the mesh.
  5. snapshot — render a PNG preview: --view iso|front|…, --section, --body NAME:opacity=…,color=#rrggbb, --width/--height.
  6. inspect — watertightness, winding, normals, non-manifold edges, boundary loops, degenerate faces, bounds, signed volume → ready_to_print / needs_repair / blocked.
  7. repair — weld, drop bad faces, fill single-face and ordered-loop holes, fix winding and inversion. Only for meshes that came from elsewhere.

The Blender half (30 skills)

skills/text-to-blender is the router for the other twenty-nine:

  • production: blender-modeling, blender-materials, blender-lighting, blender-cameras, blender-rendering, blender-animation, blender-export, blender-uv-texturing, blender-pro-workflow, blender-skill-harmonizer
  • HUD and animation QA: orbital-hud-motion, animation-quality-gate
  • reference reconstruction: reference-to-3d, wireframe-to-3d, mascot-logo-reconstruction, orthographic-registration, multiview-constraint-solver, multiview-fit-loop, fit-repair-optimizer, landmark-fit-repair, reference-analysis-validator, reference-look-calibration, source-part-segmentation, contour-to-mesh, atlas-uv-fitting, closed-surface-uv-coverage, texture-driven-mesh-fitting, texture-state-animation, quality-refinement-autoloop
  • They need a local Blender and a Blender MCP server (mcp__blender__* or mcp__blender_official__*). If neither tool set is present, say Blender is missing — never describe a render you did not make.
  • Before the first Blender call, read skills/text-to-blender/references/mcp-backends.md; keep one backend for the whole task.
  • They are upstream text (MIT, © 2026 RobLe3). Never edit them. Routing and local advice belong in skills/3d-fabric/SKILL.md and this file.
  • The Blender half and the CAD half do not share state: a Blender mesh reaches the printer only through run.py --action inspect/repair on an exported STL.

Where the command reference lives

references/solidcraft/ bundles SolidCraft's four agent reference documents (modeling, editing, checking, mechanisms) verbatim under their MIT licence. Read references/solidcraft/solidcraft-modeling/reference.md before writing steps. Never edit those files — they are upstream text; put local advice in skills/3d-fabric/SKILL.md instead.

Hard rules

  • Do not invent SolidCraft command names. Check with solidcraft-cli commands | grep <word> before writing a step, and read the bundled references/solidcraft/solidcraft-modeling/reference.md.
  • Change a parameter with parameters.change — set_parameter does not exist.
  • A run that reports success may still have written nothing; model verifies the export and turns that into FAIL. Trust the gate, not the exit code.
  • A cut that misses still returns success. Compare the reported volume with your own calculation.
  • Treat error: step N (cmd) as: command N of the list (1-based), fix that one, re-run the whole script. SolidCraft is all-or-nothing.
  • Never modify the user's original .sc; everything is written beside the output.
  • Never present a STEP export as printable: it is a B-rep with no triangles.
  • inspect and repair do not require SolidCraft. Do not refuse them when the CLI is absent.

Full skill contract: skills/3d-fabric/SKILL.md.

Download now (ZIP)

Does it work?

Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.

No reviews yet

People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.

  • It worked
  • Partly
  • It didn't
What did you use it for? e.g. Reviewing pull requests in a Rails app

Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.

Your trail

Files you open will collect here, so you can find them again.

Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.

Now do this for your own.

Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.