What does this skill do, and when should you use it?
step-parts is one of 13 skills bundled in the earthtojake/text-to-cad collection. It teaches a coding agent how to find, evaluate, and download purchasable CAD parts through the hosted step.parts API, covering off-the-shelf actuators, servos, motors, electronics boards, connectors, screws, bolts, nuts, washers, and bearings. It ships a deterministic Python downloader with fuzzy search, facet filtering, and SHA-256 checksum verification, and enforces a strict rule that network failures trigger a retry before any part is declared unavailable. When no match exists, the skill records the miss and falls back to a documented envelope rather than pretending the part was found.
- Interprets a request into search terms and optional facets (category, family, standard, tag) and queries /v1/parts
- Retries aliases, dropped letters, vendor spellings, and family facets for named actuator models like STS3215 before declaring a miss
- Downloads the selected part's stepUrl STEP file and verifies it against the record's sha256
- Runs scripts/download_step_part.py for deterministic search, download, and checksum verification with flags like --download, --tag, --limit, --all
- Returns the local path, part id, and page/API URLs so provenance is traceable
- Exposes catalog discovery endpoints (/v1/catalog/parts.index., /v1/catalog/schema, /v1/openapi.) for field semantics and client generation
- A mechanical engineer needs a real M3x12 socket head cap screw in an assembly and asks the agent to search ISO 4762 and download the official STEP file
- A robotics developer wants a Feetech STS3215 servo in their model and needs the agent to try aliases until it finds the exact part's CAD file
- A designer looking for catalog parts like a 608ZZ bearing wants facet-filtered candidates compared side by side with provenance links
- An automated agent workflow needs to batch-fetch a full page of search results as individual STEP downloads, each checksum-verified
- Users who need a self-hosted or offline parts library — the skill depends on the hosted api.step.parts catalog and cannot work without network access
- Users who want to generate custom part geometry from scratch — that is the CAD skill's job in this collection; this skill only searches and downloads catalog components
How do you install this skill?
- Core function depends entirely on the overseas-hosted api.step.parts; mainland-China reachability is undeclared and no mirror is offered, so verify connectivity yourself.
- Downloads default to the system temp directory; pass --out-dir explicitly for persistence, and manually confirm the selected part before downloading.
- Static review executed nothing: API behavior, catalog coverage, and the checksum mechanism are unverified by execution.
- The skill has no independent version number or changelog; updates require tracking the upstream repository.
- Publisher is not verified by the FollowSkills registry and is treated as unknown.
- Shell / CLI
- Network access
- Local filesystem
Python 3 (for scripts/download_step_part.py)Network access to api.step.parts
This skill ships as part of the earthtojake/text-to-cad collection (13 skills); installing the collection brings it in. The README documents no command for installing this skill alone.
Claude Code
claude plugin marketplace add earthtojake/text-to-cad#latest
claude plugin install text-to-cad@earthtojakeCodex
codex plugin marketplace add earthtojake/text-to-cad --ref latest
codex plugin add text-to-cad@earthtojakeCursor
git clone --depth 1 --branch latest https://github.com/earthtojake/text-to-cad ~/.cursor/plugins/local/text-to-cadGrok Build
grok plugin install earthtojake/text-to-cad@latest --trust
grok plugin enable text-to-cadGemini
gemini extensions install https://github.com/earthtojake/text-to-cad --ref latest --consent --auto-updateOther agents (Skills CLI)
npx skills add earthtojake/text-to-cad#latestHow do you use this skill?
Once installed, send your agent any of these to trigger it:
- Add an M3x12 socket head cap screw to my assembly — find it on step.parts and download the STEP file into the project directory
- Search step.parts for the Feetech STS3215 servo and download its STEP file with checksum verification if you find it
- Find 608ZZ bearings on step.parts and list the top five candidates with standard, name, and key dimensions
- Download the STEP file for every nut on the current results page, without overwriting any existing files
The skill is triggered by the model from the SKILL.md description: when an assembly includes a named off-the-shelf part, the agent should search step.parts before creating placeholder geometry. The flow is: parse search terms → call /v1/parts and inspect items/total/facets → present candidates when ambiguous → download stepUrl and verify sha256 → return the local path plus provenance URLs. You can also drive the bundled downloader directly:
python scripts/download_step_part.py "M3 socket head 12" --download
python scripts/download_step_part.py --id iso4762_socket_head_cap_screw_m3x12 --download
python scripts/download_step_part.py "bearing 608zz" --limit 5Useful options: --origin to override the API origin, repeatable --tag/--category/--family/--standard facet filters, --out-dir for a persistent destination (defaults to the system temp directory), --filename to rename (incompatible with --all), --limit/--page for pagination (default 10, API cap 500), --all to download every result on a page, and --overwrite to replace existing files. The script prints JSON to stdout; failures print one line to stderr and exit 1. Endpoint details live in references/step-parts-api.md.
What are this skill's strengths and limitations?
- Targets real purchasable parts and delivers usable STEP files instead of simplified stand-in geometry
- Built-in sha256 verification and a deterministic downloader make results reproducible and provenance traceable
- Thoughtful handling of fuzzy model names: aliases, vendor spellings, and family facets all have explicit retry strategies
- Clear discipline around network failures: one retry before a miss is reported, so parts are never falsely declared unavailable
- Entirely dependent on api.step.parts availability — there is no offline or local catalog fallback
- The README does not verify this individual skill on every host beyond the plugin install paths; plugin telemetry is on by default (disable with uvx cadgen telemetry off)
- STEP downloads rely on a Python script; without the plugin you must install uv and use the Skills CLI yourself
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 |
|---|---|---|---|---|
| step.parts Catalog Skill this page | 55 · Use with care | ★ 19k | 1d ago | MIT |
| FreeHire Tech Job Search Skill | 63 · Recommended | ★ 45k | 3d ago | MIT |
| AMC Sample Dataset Calibration ✓ NVIDIA · Official | 55 · Use with care | ★ 3.5k | 3d ago | Apache-2.0 |
| AutoMagicCalib Video Calibration ✓ NVIDIA · Official | 53 · Use with care | ★ 3.5k | 3d ago | Apache-2.0 |
| Xberg API Server & MCP Protocol Integration | 52 · Use with care | ★ 9.4k | 3d ago | MIT |
The sibling CAD skill in the same repository generates and edits 3D models from natural language or images (exporting STEP/STL/3MF/GLB); this skill is complementary — the CAD skill makes geometry, while this skill retrieves real off-the-shelf parts before geometry is built, so standard components don't get remodeled by hand.
How did FollowSkills review this skill?
Script hits only the user/default API origin, verifies sha256, refuses silent overwrite, treats catalog records as data not instructions, and outputs transparent provenance (stdout JSON with source URLs). Deductions: downloads default to a system temp dir without explicit confirmation, stepUrl may resolve to external GitHub LFS/Vercel Blob with checksums only as self-consistent as the record itself, publisher identity unverified, and rollback/isolation guidance is thin.
SKILL.md, API reference and script are mutually consistent; argument validation, one-line stderr failures with exit 1, adjustable timeout, and a clear alias-retry policy. Deductions: static review cannot reproduce key paths; no committed tests covering this skill path were shown; edge semantics (pagination, --all/--filename conflicts) rest on inference despite documented network-failure semantics.
Scenario (finding purchasable standard parts by name/standard and downloading STEP) is clear; trigger description is specific (aliases, standard numbers, facets) and fallback to placeholder geometry is defined. Deductions: core function depends entirely on the overseas-hosted api.step.parts with no declared mainland-China reachability or mirror; non-fit boundaries (private/custom parts) only implied.
Well-layered docs (SKILL.md → references/api.md → script), MIT license, provenance statement, and install caveats (default temp out-dir). Deductions: no skill-level version/changelog, maintenance responsibility rests on an individual repo, no FAQ or known-limitations list.
Clear value claim (avoid placeholder geometry, canonical checksummed STEP files) with marginal value over manual search and directly usable output. Deductions: static review cannot verify actual output or API behavior; benefit depends on third-party catalog coverage; no representative output evidence.
Auditable primary material exists (script, API reference, OpenAPI pointer, checksum mechanism) with facts mostly separated from inference. Deductions: below the static ceiling, no third-party execution evidence or committed tests for the skill's key paths; corroboration is thin.
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 →