What does this skill do, and when should you use it?
This skill comes from the earthtojake/text-to-cad repository (a collection of 13 skills) and is dedicated to pre-upload file review for SendCutSend.com orders. It reads your DXF or STEP/STP file, measures geometry locally with build123d and ezdxf, fetches SendCutSend's live ordering guide, catalog JSON, and specs JSON as evidence, and produces a preflight report with restrained status labels and cited rule sources. It covers laser cutting, CNC routing, bending, tapping, countersinking, hardware insertion, and finishing, and deliberately marks gaps as "need more info" rather than guessing. It suits anyone delegating their SendCutSend prototyping workflow to an agent who wants upload-blocking issues caught early.
- Inspects DXF files: units ($INSUNITS), 1:1 scale, closed profiles, duplicate or degenerate entities, bend lines with per-line local flange depths, text and annotation layers
- Inspects STEP/STP files: parseability, solid body vs. loose surfaces, bounding box, sheet thickness, cylindrical bend-face radii
- Fetches SendCutSend's official ordering guide, catalog JSON, and specs JSON live, joining records by exact SKU for material, thickness, and service availability
- Compares measured geometry against SKU bending radii, minimum flange lengths, hole sizes, bridges, and more, with ✅/❌/❓ status rows
- Proactively produces annotated diagnostic images when image generation is available; every report row cites a rule source link or "Direct file inspection"
- A mechanical engineer who wants to verify that flange depth on both sides of every bend line in a sheet-metal DXF meets the minimum for the selected material before uploading to SendCutSend
- A robotics team submitting a STEP model for CNC routing who wants confirmation it is a solid body with correct units and an in-catalog thickness
- A hardware founder ordering laser-cut enclosures repeatedly who wants the agent to auto-check hole sizes, slot widths, and secondary services (tapping, hardware insertion) each time
- A CAD beginner whose DXF export lacks $INSUNITS and who needs the scale error flagged with a re-export recommendation rather than silent rescaling
- Preflight checks for fabrication services other than SendCutSend — the skill is explicitly scoped to SendCutSend.com reports
- Users who want to generate CAD models — the repository has a separate CAD skill; this one only reviews files before upload
- Offline environments or setups that cannot reach SendCutSend's official sources — when fetches fail the skill refuses to issue ready verdicts
How do you install this skill?
- Core checks require runtime fetches from the sendcutsend.com CDN; when unreachable, the skill can only produce ❓ results. Users in mainland China should expect degraded availability.
- The skill is instruction-only documentation with no bundled tests or fixture files; check logic correctness depends entirely on model execution and is unverified.
- The runtime makes outbound HTTP requests to official CDN domains; review these domains in network policy if strict isolation is required.
- Depends on build123d/ezdxf/uv with no version constraints declared inside the skill files; measurement behavior may drift with dependency versions.
- Shell / CLI
- Network access
- Local filesystem
Python with build123d (and ezdxf)uvSendCutSend official sources via network fetch
Install the full text-to-cad collection first; this skill lives at skills/sendcutsend. Install routes by host:
Claude Code
claude plugin marketplace add earthtojake/text-to-cad#latest
claude plugin install text-to-cad@earthtojakeCursor
git clone --depth 1 --branch latest https://github.com/earthtojake/text-to-cad ~/.cursor/plugins/local/text-to-cadGemini
gemini extensions install https://github.com/earthtojake/text-to-cad --ref latest --consent --auto-updateGrok Build
grok plugin install earthtojake/text-to-cad@latest --trust
grok plugin enable text-to-cadOther agents (Skills CLI)
npx skills add earthtojake/text-to-cad#latestNote: CAD runs through uv; Windows 11's Smart App Control may block the unsigned OCP native module — disable it or run under WSL.
How do you use this skill?
Once installed, send your agent any of these to trigger it:
- Check bracket.dxf against SendCutSend laser-cut 5052 aluminum 2mm bending requirements and give me a preflight report
- Review chassis.step for a SendCutSend CNC routing upload — confirm solid body and minimum hole sizes
- My DXF has three bend lines; verify the local flange depth on both sides of each against the selected SKU minimums
- Run a SendCutSend preflight on handle.stp: units, sheet thickness, and bend radii versus the chosen SKU
After installing the collection, trigger the skill in conversation by giving a DXF or STEP/STP file path plus your order intent (process, material/SKU, thickness, quantity, secondary services). The flow: fetch official sources and record access dates, measure the actual file with build123d (STEP) or ezdxf (DXF), join catalog/spec records by exact SKU, then output a findings table with a Rule source column. If order context is ambiguous it lists candidate SKUs/materials and asks you to confirm rather than deciding for you. If official sources cannot be fetched, it will not issue ready verdicts.
What are this skill's strengths and limitations?
- Every conclusion must cite a source field path or guide section, or explicitly say "Direct file inspection" — highly auditable
- Refuses binary pass/fail when evidence is missing, conflicting, or N/A, avoiding misleading passes
- Fine-grained bend checking: per-bend-line local flange depth, bend radii, die-area crossings reported as separate findings
- Fetches official sources live each review, so rules don't go stale as SendCutSend updates its policies
- Vendor-specific: only useful for SendCutSend orders
- Requires live network fetches; no ready verdicts when official sources are unreachable
- Needs a Python/uv environment with build123d and ezdxf; the $cad skill can be used for STEP/DXF inspection but is optional
- Reports every measured risk as an error and does not mimic SendCutSend's official UI classification
How does this skill compare with similar options?
Side by side with related skills; every score comes from the same FSRS standard.
| Skill | FS score | Stars | Last updated | License |
|---|---|---|---|---|
| SendCutSend Upload Preflight this page | 57 · Use with care | ★ 19k | 1d ago | MIT |
| DXF Generation & Validation Skill | 57 · Use with care | ★ 19k | 1d ago | MIT |
| Engineering Drawing (Text-to-CAD) | 59 · Recommended | ★ 19k | 1d ago | MIT |
| DFM Review Skill | 61 · Recommended | ★ 19k | 1d ago | MIT |
| G-code Slicing Skill (text-to-cad) | 60 · Recommended | ★ 19k | 1d ago | MIT |
The same repository's DFM skill handles general sheet-metal/CNC/injection-molding review and measures draft and undercuts from a mesh, whereas this skill is bound specifically to SendCutSend's catalog, specs, and ordering guide with per-SKU reports. If you fabricate exclusively through SendCutSend, this skill's evidence chain maps more directly onto its ordering flow.
How did FollowSkills review this skill?
The skill performs local read-only geometry inspection and report generation; it mandates fetching evidence directly from official URLs, treating external data as data not instructions, and withholding verdicts when evidence is missing — good least-privilege and data-flow disclosure. Deducted for: runtime fetching of external CDN content with no supply-chain/tampering discussion; user confirmation only for ambiguous order context; rollback is moot but unaddressed, so not full marks.
Instructions are self-consistent with detailed defensive parsing rules ($INSUNITS, N/A, missing fields, conflicts, a ❓ status) and designed failure paths. Deducted for: static-only review, no committed tests or fixture files covering this skill's key paths, undeclared build123d/ezdxf version behavior, and unverified failure-message quality.
Audience and scenario are clear (SendCutSend upload preflight); the description tightly scopes triggers, and non-fit boundaries (no UI-classification inference) are explicit. Deducted for: core function depends on the sendcutsend.com CDN over overseas networks with unclear mainland-China reachability and no declared fallback; no Chinese-language support.
Well-layered docs (SKILL.md plus two references), MIT license, clear provenance, and a full report template. Deducted for: no skill-level versioning or changelog, maintenance responsibility only implicit via repo link, dependency/install notes (build123d/ezdxf/uv) live in the README rather than the skill, no FAQ or known-limitations list.
The value claim is clear and differentiated (cited-rule preflight reports versus blind upload), and the template should yield directly usable structured reports. Deducted for: static review cannot verify output correctness or direct usability; checks depend on the stability of external JSON field paths; no execution evidence.
The reporting design demands per-row source URLs, field paths, access dates and _meta metadata, with fact/inference separation. Deducted for: none of these promises are execution-reproduced here; no tests or third-party runtime evidence; claims rest on the skill's own text, so scored at the low end of the static cap.
Open a dimension to read why it scored that way
Evidence confidence：Low — Mostly static review, author material or a limited demo; useful for discovery, not high-risk decisions.
See the full review method →