solidworks-GPT-plugin

An installable Codex plugin that ports the knowledge workflow of
solidworks-claude-plugin
to OpenAI Codex.

It adds five coordinated skills:

• solidworks-design orchestrates requirements, planning, modeling, validation, and delivery.
• sw-pre-start loads conventions, active design rules, relevant references, and initial rule checks.
• sw-kb-api looks up catalog parts, published build instructions, macros, known errors, and lessons.
• sw-learner builds a complete, corrected feedback payload from the current SolidWorks session.
• sw-session-reporter asks for consent and submits feedback to the knowledge base.

_{Conceptual workflow preview—not a recorded CAD result. Every reported model still has to pass the measured validation gates shown above.}

What the workflow guarantees

This plugin is designed for engineering work where a plausible answer is not
enough. It makes requirements, knowledge lookup, modeling, and validation
explicit parts of one workflow:

flowchart LR
    A[Design request] --> B[Requirements and pre-start checks]
    B --> C[Standards and knowledge lookup]
    C --> D[Feature and datum plan]
    D --> E[SolidWorks modeling]
    E --> F{validate_document passes?}
    F -- No --> G[Inspect and repair]
    G --> E
    F -- Yes --> H[Save, export, and report evidence]
    H --> I[Build session lessons]
    I --> J{User consent?}
    J -- Yes --> K[Submit validated feedback]
    J -- No --> L[Keep the session local]

60-second tour

After installing the plugin, start a new Codex task with a concrete engineering
request. For example:

Design a metric stepped shaft in SolidWorks.
Units: mm
Material: AISI 1045
Overall length: 160
Bearing seats: 25 h6 x 30 long at both ends
Center section: 35 mm diameter
Output: saved part, STEP export, mass, bounding box, validation report

The workflow will stop for any geometry-changing requirement that is still
ambiguous, retrieve applicable standards and known patterns, plan the model,
build each document, and report measured validation evidence. A knowledge-base
outage is recorded but does not block local CAD work.

The default knowledge base is https://sw-plugin.ideep.org. Override it with
the SW_KB_HOST environment variable. Read-only runtime endpoints are public;
session feedback is sent only after the user chooses Yes, send now or has
previously chosen Always send.

Install from GitHub

codex plugin marketplace add Erfouni/solidworks-GPT-plugin
codex plugin add solidworks-gpt-plugin@solidworks-gpt

Restart the ChatGPT desktop app or start a new Codex task after installation so
the new plugin skills are loaded.

For a local checkout, replace the GitHub shorthand in the first command with
the absolute path to this repository.

Requirements

• Windows with SolidWorks installed for actual CAD work.
• A SolidWorks automation surface already available to Codex (for example a
  SolidWorks MCP/COM integration). This plugin supplies the design and knowledge
  workflow; it does not bundle an MCP server or install SolidWorks itself.
• curl for knowledge-base calls.
• Python 3.9+ for the optional deterministic session and feedback utilities.

Knowledge-base outages never block CAD work. The skills record the outage and
continue with the available SolidWorks tooling and engineering context.

Diagnose local readiness

Run the bundled doctor before the first CAD task:

python plugins/solidworks-gpt-plugin/scripts/doctor.py

It validates Python, the installed manifest/skills/schema, SW_KB_HOST syntax,
curl, and the local SolidWorks COM registration. It makes zero network
requests and never launches SolidWorks. Missing optional dependencies are
reported as warnings; require the complete CAD and feedback environment with:

python plugins/solidworks-gpt-plugin/scripts/doctor.py --strict

For CI, support packets, or reproducible installation evidence, use --json.
Host output is credential-free, and endpoint reachability is deliberately not
claimed.

For release notes and project questions, see the
latest release,
the changelog, and
GitHub Discussions.

Privacy and local state

• .sw-learner-state.json stores the current session ID and submission metadata
  in the working directory.
• ~/.sw-feedback-pref contains always only after the user explicitly chooses
  Always send.
• Feedback may include build instructions, generated macros, resolved errors,
  lessons, and model images. The payload is never submitted without consent
  unless the saved always preference is present.

Validate the plugin

From this repository root:

python plugins/solidworks-gpt-plugin/scripts/run_checks.py

The repository also includes the original porting specification and the KB
admin OpenAPI reference under plugins/solidworks-gpt-plugin/docs/.

Contributing

Bug reports, documentation improvements, and focused pull requests are
welcome. Start with CONTRIBUTING.md, use the structured issue
templates, and run the repository checks before opening a pull request.

License

MIT. See LICENSE and NOTICE for upstream attribution.