---

name: cortex-m description: "Develop, export, build, and test ExecuTorch Cortex-M/CMSIS-NN workflows with PT2E quantization, graph rewrites, dialect tests, and Arm toolchain prerequisites." disable-model-invocation: true metadata: disco-role: operating license: NOASSERTION

cortex-m

Use this sub-skill for ExecuTorch Cortex-M, CMSIS-NN, Arm MCU, Corstone/FVP, bare-metal, or Cortex-M quantizer/pass-manager tasks.

What This Owns

• PT2E quantization with CortexMQuantizer.
• Graph rewrite flow using CortexMPassManager and Cortex-M custom ops.
• Distinguishing pure Python/dialect graph tests from implementation tests that require Arm tools/FVP.
• Bare-metal runner planning and troubleshooting.

Fast Path

1. Decide whether the user is asking for graph/export correctness or hardware implementation correctness.
2. For graph/export, use API reference and plan dialect tests first.
3. For implementation, confirm Arm toolchain/FVP and any license/EULA prerequisites before running commands.
4. Use the planner:

python scripts/plan_cortex_m_test.py --mode dialect
python scripts/plan_cortex_m_test.py --mode implementation --model conv2d

Cross-Links

• Generic source build issues: ../setup-build/SKILL.md.
• Backend choice and Arm Ethos-U/VGF alternatives: ../backend-selection/SKILL.md.
• Export runtime sequence outside Cortex-M-specific passes: ../export-runtime/SKILL.md.