Skills: requirements and platform support

The skills overview lists what each skill does. This page lists what each skill needs on your machine, and which AI tools can run it in full.

Requirements

Skill	Requirements
qt-cpp-review	Python 3.7+ for the linter script. No external Python packages.
qt-qml-review	Python 3.7+ for the linter script. No external Python packages. Optional: qmllint from a Qt 6 installation for type checks.
qt-qml-profiler	A Qt 6 installation containing bin/qmlprofiler. CMake and a C++ toolchain (profiling mode only, to build with -DQT_QML_DEBUG). Python 3 for the trace parser.
qt-qml-test	Qt 6: the generated tests use unversioned import QtQuick and import QtTest. Existing QML components to test; the skill does not scaffold a project from nothing.
qt-qml-test-run	A Qt 6 installation containing bin/qmltestrunner. A CMake-based project (qmake is not supported). CMake 3.21 or newer and a C++ toolchain when building. Python 3 for the JUnit XML parser.
qt-cmake-project	Qt 6.8 or newer is a soft requirement: the skill's defaults assume the Qt 6.8 API. No script dependencies.
qt-figma-token-extraction	A Figma file with variables or text styles. The Figma MCP server, or a Figma Personal Access Token for REST API extraction. An existing CMake-based Qt 6 project, or agreement to create one.
qt-figma-component-generation	The Figma MCP server. design-tokens.json and the QML design-system singletons produced by qt-figma-token-extraction.
qt-cpp-docs, qt-qml-docs, qt-ui-design	No external dependencies.

For background on what qmlprofiler measures and how each event type is interpreted, see Profiling QML applications in the Qt Creator documentation.

Platform support

The review skills (qt-cpp-review, qt-qml-review) combine a Python linter with parallel subagents. The full workflow runs on Claude Code and Codex CLI. GitHub Copilot reads the same skill directory, but it cannot run the linter or launch subagents, so only the static guidance loads.

qt-cpp-review and qt-qml-review launch deep-analysis agents only for scopes above about 2,500 and 1,500 lines respectively; smaller scopes are reviewed inline. They prefer the dedicated cpp-review-agent and qml-review-agent, which the plugin install ships in its agents/ directory. That directory sits outside the skill, so a skill-only copy or symlink does not include it. The skill then uses the tool's general-purpose agent and passes the same rules in the prompt, so the review works but costs a little more. On Claude Code you can add the dedicated agents by copying agents/cpp-review-agent.md and agents/qml-review-agent.md to ~/.claude/agents/.

The condensed platforms/windsurf.md variants (for qt-cpp-review, qt-qml-review, qt-qml-docs and qt-qml) carry the highest-priority rules only. Expect less thorough results than from the full skill.

qt-qml-profiler and qt-qml-test-run build and run binaries, execute Python scripts and write report files. They need an agent with shell access, such as Claude Code or Codex CLI. There are no condensed variants, and GitHub Copilot and in-IDE assistants without a build environment are not supported.

See Installation for where each tool expects the skill files.

Report accumulation in qt-qml-test-run

Every normal run writes a timestamped build/tests/reports/test-report-*.md and build/tests/reports/junit/qmltests-*.xml. The skill never rotates or deletes them, because later runs use the JUnit XML as their prior-run baseline. Over time build/tests/reports/ grows.

To limit this, pass --no-report on iterative runs. That skips the Markdown report but keeps the small JUnit XML files, so the baseline still works. You can also delete old reports by hand.