PipingToolset for ChatGPT and Codex — B31 Piping Skills for OpenAI's Assistants

Version 0.6.2 · released 2026-10-28 · Apache-2.0 · free · Content last reviewed 2026-10-28 · page regenerated 2026-10-28 at build.

B31 calcs you can reproduce and defend in review, in ChatGPT or Codex. The assistant does the interview. A script does the math. You read the table. The same seven skills that ship for Claude, from the people who build the free calculators on this site, written and maintained by a practicing, licensed piping engineer.

One plugin, either assistant

The plugin folder carries two manifests side by side: the Claude one and an Agent Plugins 1.0 one with OpenAI's listing block. Same skills, same scripts, same tests, same mark in the plugin list. Pick the assistant your team already pays for.

Watch the demo

Seven and a half minutes, recorded in Claude Cowork: the install, a wall thickness check with the hand check beside it, the refusal, and a pipe support checked and optimized from CAESAR II loads. The ChatGPT and Codex runs produce the same tables, because the same scripts produce them.

▶ Piping Toolset for Claude: B31.3 wall thickness and a pipe support check, every number from a script (7:37) — Install in Cowork, a B31.3 wall thickness check with the hand check beside it, the plugin refusing to quote an allowable stress, and a pipe support checked and optimized from CAESAR II restraint loads with the calc report written to disk. Watch on YouTube.

Use the pressure design skill. B31.3-2022, normal fluid service. NPS 6 seamless pipe, OD 6.625 in,
design 750 psig at 600 F, corrosion allowance 0.0625 in, 12.5% mill tolerance. My allowable S at
600 F is 17900 psi from my copy, E = 1.0, W = 1.0, Y = 0.4. What minimum thickness do I need and
does 0.280 in nominal wall work?

t  = 0.13650 in   B31.3 ¶304.1.2 Eq. (3a): PD / 2(SEW + PY) = 750 × 6.625 / 2(17900 + 300)
tm = 0.19900 in   t + c = 0.13650 + 0.0625
0.280 × 0.875 = 0.245 in ≥ 0.19900 in   ADEQUATE
S, E, W, Y: user-supplied. Edition: B31.3-2022 (stated).

What it does

PipingToolset is a plugin: seven skills, each a set of written instructions the assistant follows plus a standard-library Python script that performs the calculation. The assistant asks for the code basis and the inputs, writes a small specification file, runs the script by path, and reads the result back to you. The same engine behind the free calculators on this site produces the numbers.

How it works

Three rules hold in every skill, in every host:

• The assistant never does the arithmetic. Every number a skill states comes from a script run on inputs you supplied. If you see a number and no script ran, treat the answer as untrusted and ask again.
• The assistant never recalls a code table value. Allowable stresses, moduli, SIFs, spans, ASCE 7 coefficients, anchor bond values, flange ratings, gasket factors: the skill names the table, row and edition and asks you to read it from your own copy. Asking for "just the number" is a pinned test, and the answer is no.
• The edition is an input, never a silent default. Where a rule changed between editions (for example the f-factor form before and after B31.3-2022, or Appendix D giving way to B31J from B31.3-2020), the skill asks which one governs your project.

What it will not do

• It will not give you a tabulated code value. Not an allowable stress, not a quality factor, not a Y coefficient, not an SIF, not a support span, not a catalog bond strength, not a B16.5 rating, not a gasket factor, not an API nozzle allowable. You read those from the document you are licensed to use; the plugin contains none of them.
• It will not replace a flexibility analysis. Stress screening is a hand check of what the program applied. CAESAR II and AutoPIPE remain the analysis tools; the load case builder and iso-to-PCF skills exist to feed them.
• It will not pick your code or edition for you. Basis items you do not answer are marked ASSUMED on the calc sheet, never silently filled in.
• It will not compute in the model's head. A skill that cannot run its script, for example in a chat surface with no sandbox, stops and says so instead of estimating.
• It will not seal anything. Results are engineering aids. The engineer of record verifies them before issue, and you remain responsible for the result.
• It will not reproduce code text. Paragraph numbers and equation references are cited; the wording of the standard is not included.
• It will not touch your Piping Toolset account. The plugin is free and separate from the site's paid tiers. Installing it unlocks nothing on the site, and a subscription changes nothing in the plugin.

Where your data goes

Short answer: nothing you enter reaches PipingToolset or Flexible Piping Solutions. The plugin makes no network calls, carries no telemetry, needs no API key and no account with us. The only contact with our infrastructure is the download, and that is served by GitHub.

Install

In the ChatGPT desktop app and Codex

1. With the Codex CLI installed, register the marketplace once:

   codex plugin marketplace add pipingtoolset/pipingtoolset-claude

2. Install PipingToolset from the plugin picker: /plugins in the Codex CLI or IDE extension, or Customize › Plugins in the ChatGPT desktop app. It shows with the PipingToolset mark and "By Flexible Piping Solutions LLC"; the seven skills are listed beneath it.
3. Start a new conversation. Call a skill as $pressure-design (or pick it from /skills) in Codex, or name it in plain words in ChatGPT: "use the pipe support skill to check PS-201". Naming it matters, because other installed skills on the same topics would otherwise compete.

To update, run codex plugin marketplace upgrade pipingtoolset and reinstall from the picker.

From the ChatGPT plugin directory

The directory listing is submitted from the same archive and is pending review. Until it appears, the marketplace route above is the install.

Skills only, for other assistants

The skills are plain Agent Skills, so the same repository installs into Cursor, GitHub Copilot, Gemini CLI and others with npx skills add pipingtoolset/pipingtoolset-claude or gh skill install pipingtoolset/pipingtoolset-claude. For Claude Code and Claude Cowork, see the Claude page.

Requirements

Python 3.9 or newer wherever the scripts run: your own machine for the Codex CLI and IDE extension; the ChatGPT desktop app and Codex cloud provide their own sandbox. There is nothing else to install.

Try it in twenty seconds

Paste the worked example at the top of this page. Then ask for what it will not give you:

Use the pressure design skill. What's the allowable stress for A106 Grade B at 650 F under B31.3?
Just give me the number.

Expected: no number. It names B31.3 Table A-1 (A-1C in the 2024 edition), the material row and the two temperature columns, and offers to interpolate once you have read both values. Twelve more prompts with pass criteria are in the repository's test prompt set.

How it is verified

• One engine. Every script that computes a code equation reproduces the locked fixture cases of this site's calculation engine, value for value. A red parity test blocks a release. The fixtures and the method are described on the validation page and listed on the coverage page.
• Same archive, both directories. The file submitted to OpenAI is byte for byte the file Cowork installs; a test holds the OpenAI listing block to the submission portal's limits, and the agentskills.io reference validator passes every skill folder in CI.
• Golden outputs. The load case generator's tables and the PCF builder's output are locked byte for byte for the worked examples. A change to case order, combination strings or QC text is a deliberate, reviewed golden update.
• Content guard. An automated test fails the build if any skill text or script embeds a tabulated code value or third-party proprietary material.
• Recall channel. An installed copy keeps any defect it shipped with until you update it. A release labelled UNCONSERVATIVE-FIX in the changelog and on GitHub Releases corrects a result that could have been less conservative than the code requires, and names the skill, the affected inputs and the fix. Update when you see one.

Questions

The plugin is an engineering aid. Results must be independently verified and reviewed by a qualified, licensed engineer before any design, fabrication or operation decision. ChatGPT and Codex are products of OpenAI; Claude is a product of Anthropic, PBC. This plugin is not affiliated with or endorsed by either.