Thermal Machinery AI Skills

Give Codex and other AI agents an engineering workflow for building and validating gas-turbine and thermal-machinery models in MATLAB/Simulink.

Design point → steady-state closure → dynamic initialization → Simulink implementation → validation

English | 简体中文

Two open skills turn thermal-modeling knowledge into staged, auditable work: agents establish the physical model before writing a block diagram or tuning a controller.

Install and try it

The following commands were verified against the public GitHub repository with the skills CLI: it discovered both published skills by name and did not install anything during the check.

npx skills add YYch89/thermal-machinery-ai-skills --list
npx skills add YYch89/thermal-machinery-ai-skills --skill gas-turbine-ai-modeling

To install the broader thermal-system skill instead, replace the final skill name with thermal-machinery-dynamic-modeling. The installer supports other agents as well; use its --agent option when applicable.

Manual installation remains the dependable fallback: clone the repository and copy one or both folders from skills/ into the skills directory used by your agent. Each folder is self-contained and begins with SKILL.md.

git clone https://github.com/YYch89/thermal-machinery-ai-skills.git

Then invoke a skill in your agent:

Use the gas-turbine-ai-modeling skill to build a staged design-point-to-dynamic MATLAB/Simulink workflow. Do not jump directly to Simulink: produce state tables, loop contracts, an initial-condition registry, and validation gates.

For Codex environments that load plugins, this repository also includes a validated .codex-plugin/plugin.json manifest pointing to its two existing skills.

What an agent can do with these skills

• Build a gas-turbine model from design point through steady state, rotor/volume/combustor dynamics, and MATLAB-to-Simulink staging.
• Audit compressor/turbine maps, initialization residuals, control authority, units, and physical constraints.
• Plan broader Brayton, Rankine, refrigeration, heat-pump, and hybrid thermal systems through topology and stream ledgers.
• Repair or review an existing dynamic Simulink model without treating a successful run as proof of validation.

Why use a skill instead of a prompt?

A direct request such as Build me a dynamic gas-turbine Simulink model can cause an agent to jump straight to implementation. These skills require an explicit engineering sequence instead:

Requirements → topology → design point → steady closure → initialization → dynamics → control → validation

Not just prompts: a staged engineering workflow. The output is expected to include auditable artifacts, assumptions, balances, initial conditions, and open validation gates—not only generated code.

60-second example: a synthetic dual-shaft gas turbine

The public example at examples/synthetic-dual-shaft-gt-dynamic demonstrates the workflow with synthetic data:

Input	Skill workflow	Output
Build a dynamic dual-shaft gas-turbine generator model.	Design-point state data → component contracts → rated steady closure → initial-condition residual check → rotor and volume dynamics → validation gates.	Runnable MATLAB/Simulink scripts and tests for an exploratory, reduced model.

From the example root, MATLAB users can run:

runtests('tests')
run('scripts/build_gt_simulink_model.m')
run('scripts/run_simulink_closed_loop.m')
run('scripts/build_gt_component_native_model.m')
run('scripts/run_component_native_closed_loop.m')

The generated .slx files are deliberately not committed. See the example README for its model boundary, assumptions, and expected checks.

Choose a skill

gas-turbine-ai-modeling

Use for gas-turbine design points, off-design matching and maps, rotor/volume/combustor dynamics, MATLAB/Simulink staging, initialization, control integration, and model debugging or repair. It is especially suited to multi-shaft gas turbines.

thermal-machinery-dynamic-modeling

Use for wider thermal machinery: heat engines, hybrid thermal systems, heat pumps, Brayton/Rankine/refrigeration systems, and thermo-fluid plants where topology, stream ledgers, modeling depth, and validation must be made explicit.

For an integrated system, use the general skill for the overall topology and stream ledger, then the gas-turbine skill for the gas-turbine subsystem.

Reproducibility and scope

• The included dual-shaft gas-turbine case is synthetic, public, exploratory, and reduced; it is not a manufacturer or certified-engine model.

• Runtime smoke checks for the public example were previously verified with MATLAB/Simulink R2023a:

  R2023_PUBLIC_GT_CORE_CHECKS_PASSED
  R2023_PUBLIC_GT_RUNTTESTS_PASSED
  R2023_PUBLIC_GT_README_SCRIPTS_COMPLETED
  

• The skills require source-backed equations, component data, stated assumptions, residual checks, and domain review before an engineering claim is made.

• The repository excludes proprietary maps, private calibration data, unpublished project models, and safety-critical certification artifacts.

The compact synthetic-heat-pump-ledger example shows the general workflow’s topology, node-ledger, and validation boundaries outside the gas-turbine domain.

Citation

If you use this project, please cite:

Wen, J. (2026). Thermal Machinery AI Skills (v0.1.1) [Software]. Zenodo. https://doi.org/10.5281/zenodo.20666571

Development and release checks

python scripts/validate_release.py

This verifies repository structure, privacy, and packaging only; it does not validate engineering correctness. See docs/release-checklist.md and docs/external-review-prompts.md for the release and review material.

License

MIT License.