Known limitations

These limitations apply to the beta. Verify generated projects, build output, device operations, and agent responses before relying on them. This list was reviewed on November 4, 2026.

Limitation	Affected component/version	Workaround or notes
STM32CubeMX2 GUI IP errors and warnings are not propagated by the connector.	STM32CubeMX2 Context.	Check STM32CubeMX2 GUI diagnostics before generation. A successful connector readiness result does not prove that the GUI has no blocking IP errors.
STM32CubeMX2 can display EPIPE: broken pipe because of local shell settings.	Local terminal environment used to launch STM32CubeMX2.	Check the shell environment before treating the popup as a project configuration failure. See Why does MX2 fail in a terminal? in the FAQ.
The toolkit is for evaluation and integration validation. No target is production-validated.	All contexts in this beta.	Verify agent output, project files, generated code, and device state. Keep backups and follow the tools’ normal safety procedures.
Build and debug require a VS Code project setup.	Build and DAP Debug contexts. See Supported contexts and versions for current versions.	These contexts need the VS Code target. Follow the Build and Debug guides for setup and tool requirements.
HAL1-to-HAL2 project migration is not delivered.	STM32Cube Embedded Software Context.	Use the delivered example discovery, HAL/LL guidance, and application-code services. The toolkit does not provide a project-migration workflow in this release.
ST Docs does not yet support board user manuals.	ST Official Documentation Context. See Supported contexts and versions for current versions.	Search the Sidekick Knowledge Base for board manuals, including the NUCLEO-C562RE user manual. For other missing official documents, check st.com.
Accepted target combinations are not all integration-tested.	Installer target catalog. See Supported IDEs and AI coding agents.	VS Code with GitHub Copilot and Claude Code CLI are integration-tested pairings, not fully validated workflows. Cursor setups and other Claude Code setups are only partially validated. Other supported combinations are checked for file placement. Verify agent behavior in the selected host.
An AI agent’s shell tool may not support interactive installation questions.	Observed with Claude Code in VS Code on Windows. Behavior depends on the agent’s shell tool and configuration.	Run the interactive deployment step yourself in a normal terminal, then let the agent continue with dependency repair and verification. See Interactive installation through an AI agent’s shell.
VS Code 1.142.x is excluded from this beta because of an upstream Copilot Local-mode change.	Beta host compatibility, not a connector version.	VS Code 1.141.x and 1.143.x are supported. Avoid 1.142.x. See microsoft/vscode#341937.

For the reason the toolkit is in beta, see Why is STM32Cube for Agentic AI released as a beta?. For changes to the current context inventory, see Supported contexts and versions.

Interactive installation through an AI agent’s shell

Custom installation asks which contexts and services to deploy. Some AI agent shell tools can launch a command but cannot send answers to its standard input (stdin) while it runs. The command may wait at its first question, exit without a deployment result, or report an input error. This behavior was observed with Claude Code in VS Code on Windows. It is not a limitation of all Claude Code versions or configurations, nor of the VS Code integrated terminal itself.

A background task does not necessarily provide interactive input. Starting the command again, piping a sequence of answers, or using a named pipe is not a reliable substitute for a supported interactive session.

If the agent cannot relay the questions, run the custom installation’s project-deploy-installed command yourself in a normal terminal, such as the VS Code integrated terminal. Keep the selected project folder and target options, and answer each context and service question there. If an earlier invocation is still waiting, cancel that invocation before retrying. Give the agent the final command result so it can continue with doctor --fix, using the selected VS Code profile where applicable, and the first-use checks.

Do not treat a displayed question or an exit without a deployment result as successful installation. The agent must not replace your service choices with defaults or deploy all contexts unless you explicitly request that selection. For target-specific commands, follow the custom installation procedure in the installation guide linked in SEE ALSO.