What does this skill do, and when should you use it?
dfm is one of 13 skills bundled in the earthtojake/text-to-cad repository. It performs guided, process-specific DFM reviews of a single part across sheet metal (bends, reliefs, flat patterns), CNC machining (tool access, internal corners, deep features, setups) and injection molding (draft, undercuts, projected area). It is explicitly not an automatic feature-recognition or manufacturing certification engine: it picks the process first, walks a per-process checklist, and reports unavailable checks rather than guessing. Injection molding gets a measurement script, mold_tool.py, that reports draft, undercuts and projected area from a mesh as facts only; the other two processes rely on CAD inspection or supplied dimensions. Every numerical finding must cite units, measurement method and source section, and conclusions are limited to pass, fail, review or unverified.
- Reads the part (STEP/B-rep preferred, mesh or screenshot as fallback) and identifies the process: sheet metal, CNC machining/turning or injection molding, possibly several at once for comparison
- Reads the matching reference checklist (sheet-metal.md, cnc.md or injection-molding.md) for the selected process
- For injection molding, runs scripts/mold_tool.py to measure draft, undercuts and projected area (fact-only output, never pass/fail)
- Prefers the user's actual supplier/tooling specification over general guidance, recording source version, date, section and conditions for every adopted limit
- Produces a structured Markdown report with scope, per-finding evidence/rule/result/suggested change, and coverage of checks not performed
- For redesign requests, edits the source via the $cad skill when available and rechecks affected neighboring features
- A mechanical engineer checking bends, reliefs and flat patterns on a sheet metal part before sending it to a shop
- A CNC designer verifying internal corner radii, deep cavities and tool access against a specific shop's spec
- An injection-molded part designer needing measured draft angles and undercut locations to decide about side actions
- A buyer comparing manufacturability of the same part across processes (e.g. machining vs molding)
- Teams that need a formal DFM report with evidence and rule provenance rather than an opinion
- Anyone expecting automatic feature recognition or a one-click manufacturability certification — the skill explicitly disclaims both
- Additive manufacturing printability — that is handled by the separate dfam-check skill in the same repo, not this one
- Users with no usable geometry (only a render and no dimensions) — most checks will end up review/unverified rather than measured verdicts
How do you install this skill?
- Static review only; the script was not executed, so claimed measurement precision (5th-percentile curved-face reads, raster resolution) is not independently reproduced.
- Installing the text-to-cad bundle enables telemetry by default; disable with `uvx cadgen telemetry off` or CADGEN_TELEMETRY=0 if not wanted.
- The measurement script depends on trimesh/numpy/rtree via requirements.txt, whose contents are not shown in the evidence.
- All process limits come from supplier pages and may change; verify thresholds against current specs rather than treating them as universal physical limits.
- Overseas reference sites (Hubs/Protolabs) may be unreliable from mainland China, though they are optional references and do not affect local script execution.
- The skill has no independent version or changelog; updates ride on the repository's overall release cadence.
- Shell / CLI
- Network access
- Local filesystem
Python 3 (for scripts/mold_tool.py)requirements.txt dependenciesuv (via the text-to-cad plugin install)
This skill is one of 13 in the earthtojake/text-to-cad monorepo; installing the library gives you skills/dfm.
Claude Code (plugin, includes the skills and the CAD server):
claude plugin marketplace add earthtojake/text-to-cad#latest
claude plugin install text-to-cad@earthtojakeSkills CLI (any agent supporting the skills framework):
npx skills add earthtojake/text-to-cad#latestAfter installing, run requirements.txt install in the skill directory if you need the injection-molding measurement script.
How do you use this skill?
The skill triggers when a user asks whether a part can be bent, machined or molded, asks about manufacturability or tooling, or requests a DFM review or redesign. Flow: pick the process (more than one when comparing), read the matching reference file; prefer the user's supplier/tooling spec, falling back to the two sources named in the process file while recording version and access date; use the $cad skill's inspection workflow for geometry facts when available, otherwise use dimensions with provenance or mark the check unverified. For injection molding, run the measurement script:
python scripts/mold_tool.py measure part.stl --pull z
python scripts/mold_tool.py pulls part.stlReport verdicts are limited to four values: pass (measured feature meets a cited applicable limit), fail (measured violation), review (qualitative risk), unverified (missing evidence or context); if uncertainty straddles a threshold the check stays unverified. A review request alone does not request edits, uploads, ordering or machine operation.
What are this skill's strengths and limitations?
- Complete per-process checklists covering the three most common conventional processes: sheet metal, CNC and injection molding
- Evidence-first: every verdict requires a measured value, cited rule, units and applicable conditions — never inferences from renders or colors
- Clean separation of fact and judgment: the measurement script reports numbers only, pass/fail comparison belongs to the review
- Honest handling of uncertainty — unverified when evidence or rule provenance is missing, no invented feature IDs or sample values
- MIT licensed, and installable across Claude Code, Codex, Cursor, Gemini and Grok via the text-to-cad library
- Sheet metal and CNC have no geometry analyzer at all — they depend entirely on CAD inspection or user-supplied dimensions, so evidence gaps leave conclusions unverified
- It is a guided review, not automatic feature recognition; output quality depends on the geometry and supplier specs the user provides
- On Windows 11 with Smart App Control enabled, the CAD kernel (OCP) behind the whole text-to-cad library fails to load
- Reports are chat Markdown by default; wiring in a specific shop's rules requires the user to supply the vendor documentation
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 |
|---|---|---|---|---|
| DFM Review Skill this page | 61 · Recommended | ★ 19k | 1d ago | MIT |
| SendCutSend Upload Preflight | 57 · Use with care | ★ 19k | 1d ago | MIT |
| DXF Generation & Validation Skill | 57 · Use with care | ★ 19k | 1d ago | MIT |
| OWASP Security Skill | 61 · Recommended | ★ 378 | 16d ago | MIT |
| Maintainer Response (agent-service-toolkit) | 69 · Recommended | ★ 4.5k | 7d ago | MIT |
The sibling Dfam Check skill (skills/dfam-check) is the clear complementary alternative: it targets additive manufacturing, measuring mesh printability per process (wall thickness, overhangs, support volume, build orientation), while dfm covers only subtractive/forming processes (sheet metal, CNC, injection molding). Choose dfam-check for print feasibility and dfm for machining or tooling feasibility.
How did FollowSkills review this skill?
Read-only guided-review workflow; the script explicitly emits facts only, never verdicts; supplier pages are scoped as reference data, never instructions, showing strong prompt-injection hygiene; fabricated measurements are explicitly forbidden. Deducted for: repository-level default-on telemetry (disclosed and opt-out, but an accompanying data flow requiring user action to disable) and unaudited external dependencies (trimesh/rtree).
Error handling is mature: exit codes 1/2 separate hard failure from partial reports, partial flags, units-suspect signaling, and documented failure modes for mesh and curved-face pooling, plus NaN cleanup. But nothing is executed in this static read, requirements.txt is not shown in evidence, and no committed tests for this skill are displayed, capping at the anchor of 10.
Trigger description is precise (bend/machinability/moldability questions, DFM review requests); boundaries are well declared: not an automatic feature-recognition or certification engine, render-only inputs yield review items only, silicone/overmolding out of scope. Deducted for: Python/pip environment dependency, overseas reference sites (Hubs/Protolabs, optional only), and no Chinese-language support.
Well-layered docs: main SKILL.md, three process references with checklists and worked examples, agents/openai.yaml interface metadata; MIT license and provenance statement present. Deducted for: no per-skill version or changelog, the script file is truncated in evidence so it cannot be fully verified, and maintenance responsibility is only implied by a provenance link.
Clear marginal value: a fact-only measurement script plus guided review with cited rule sources beats unaided model judgment on manufacturability; the pass/fail/review/unverified report format is directly usable. However, static review cannot confirm outputs are directly usable on real artifacts, and measurement quality (facet pooling, raster precision) is not independently reproduced, so score is capped and further reduced.
Auditable primary material exists: full script source, detailed failure-case data in comments (measured fixture numbers), and a test CI badge in README. But no committed tests covering the dfm skill's key paths are shown, worked-example numbers are explicitly illustrative, and third-party execution evidence is insufficient for a higher score.
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 →