Official agentic skills for Qt software development and quality assurance, designed for use with AI coding tools such as Claude Code, Codex CLI, and GitHub Copilot.

Skills have been tested with frontier LLMs from the Claude, Gemini, and GPT model families.

There is no settled industry standard for AI skill packaging. Each platform has its own conventions. Our canonical format uses SKILL.md with YAML frontmatter in a directory-based structure — this works natively on Claude Code, Codex CLI, and GitHub Copilot, and can be adapted to other platforms through condensed variants. See CONTRIBUTING.md for the full cross-platform story.

These agentic development skills use AI and can make mistakes. Always double-check the output carefully.

Before using the skills under Qt commercial licensing, make sure you have understood and agree to the Qt AI Services Terms & Conditions. By using the skills or MCP tools, you accept these terms & conditions and that you have the right to do so on behalf of your employer.

• Review — structured code review workflows combining deterministic linters with deep AI analysis
• Process — workflows and decision frameworks (architecture, build, test, documentation)
• Conceptual — mental model corrections for areas where LLMs consistently fail (declarative QML, C++/QML boundary, Widgets patterns, UI design)
• Tool — guidance on Qt CLI tools and testing solutions

See mcp/qt-documentation-mcp/README.md for endpoint and manual setup instructions for AI clients other than Claude Code.

Read the Introducing the QML Live Preview MCP Tool for Agentic Development of Embedded Devices blog post to gain more information about the QML-Live-Preview-MCP-Server.

See mcp/qml-live-preview-mcp/README.md for endpoint and manual setup instructions.

skills/                           # All skills live here
  qt-cpp-review/                  #   Each skill is a directory
    SKILL.md                      #   with a SKILL.md entry point
    references/                   #   and optional reference docs
      lint-scripts/
      qt-review-checklist.md
    platforms/                    #   Platform-specific variants
  qt-qml-review/
  qt-qml/
  qt-canvas2d/
  qt-ui-design/
  qt-qml-docs/
  qt-cpp-docs/
  qt-qml-profiler/
  qt-qml-test/
  qt-qml-test-run/
  qt-cmake-project/
mcp/                              # MCP servers bundled with the plugin
  qt-documentation-mcp/           #   Each server is a directory
    README.md                     #   with its own README
.mcp.json                         # MCP registration for Claude Code
.claude-plugin/                   # Claude Code CLI & Copilot CLI plugin config
gemini-extension.json             # Gemini CLI extension manifest
docs/                             # Source for the docs site (MkDocs)
CONTRIBUTING.md
LICENSE
README.md

Every skill is a directory containing a SKILL.md file with YAML frontmatter. This format aligns with what Claude Code and Codex CLI support natively:

---
name: qt-qml-review
description: >-
  Reviews QML source files for correctness, performance, and
  maintainability. Deterministic linting (48+ rules) plus
  deep analysis (parallel shard agents for large scopes) for
  bindings, layout, loaders, delegates, states, and performance.
license: LicenseRef-Qt-Commercial OR BSD-3-Clause
compatibility: >-
  Designed for Claude Code, GitHub Copilot, and similar agents.
metadata:
  author: qt-ai-skills
  version: "1.0"
  qt-version: "6.x"
---

• name must be lowercase alphanumeric + hyphens, max 64 chars, and must match the directory name
• description must describe what the skill does AND when to use it — front-load the key information in the first 250 characters, as some platforms truncate beyond that point
• SKILL.md body should stay under 500 lines — move detailed content to references/ files
• Progressive disclosure: agents load only name/description at startup, read SKILL.md body on activation, and pull references/ files only when needed

Skills are authored to be tool-agnostic wherever possible. However, different AI coding tools consume skills in different ways:

Most skills work across tools without changes. When a skill needs to reach platforms that cannot read multi-file directories (e.g. Windsurf, Amazon Q), create variants in a platforms/ directory:

skills/qt-qml-review/
├── SKILL.md                       # Core skill (tool-agnostic)
├── references/
│   ├── qml-lint-rules.md
│   └── qml-review-checklist.md
└── platforms/                     # Platform-specific variants
    └── windsurf.md                #   Self-contained compact variant

The platforms/ directory is a convention for this repository. See CONTRIBUTING.md for full details on creating platform variants.

Condensed variants and tools that can't run scripts or subagents give less thorough results. For example, GitHub Copilot loads only the static guidance of the review skills, without the linter or the deep-analysis agents. See Requirements and platform support for per-skill requirements and limits.

Most AI coding tools can install skills directly from this GitHub repository. Choose the method for your platform below.

Install as a plugin using the Claude Code CLI plugin system:

/plugin marketplace add TheQtCompanyRnD/agent-skills
/plugin install qt-development-skills

Or install individual skills manually:

# Symlink a skill into your personal skills directory
ln -s "$(pwd)/skills/qt-qml-review" ~/.claude/skills/qt-qml-review

# Or copy into a project for team use
cp -r skills/qt-qml-review .claude/skills/qt-qml-review

# Review skills only: also install their review agents. The plugin
# install does this for you; a skill-only copy does not include agents/.
mkdir -p ~/.claude/agents
cp agents/cpp-review-agent.md agents/qml-review-agent.md ~/.claude/agents/

Without the agents, qt-cpp-review and qt-qml-review still work: they fall back to the general-purpose agent.

Claude Code auto-discovers skills — no restart needed.

Use Vercel's cross-platform skill installer:

npx skills add TheQtCompanyRnD/agent-skills

Or copy skills manually:

cp -r skills/qt-qml-review ~/.codex/skills/qt-qml-review

Restart Codex after adding skills.

Note: Codex is evolving rapidly. If ~/.codex/skills/ does not work, try ~/.agents/skills/ or .agents/skills/ in your project root.

Copilot natively reads the same SKILL.md directory format as Claude Code. Install individual skills using GitHub CLI (preview):

gh skill install TheQtCompanyRnD/agent-skills qt-qml-review

Or copy/symlink the skill into one of Copilot's discovery paths:

# User-wide (available in any project)
cp -r skills/qt-qml-review ~/.copilot/skills/qt-qml-review

# Or scoped to a specific project
cp -r skills/qt-qml-review .github/skills/qt-qml-review

Copilot also auto-discovers skills installed for Claude Code under .claude/skills/, so a single install can serve both tools. Use /skills list inside Copilot CLI to confirm what loaded.

qt-qml-test also ships a self-contained Copilot variant in skills/qt-qml-test/platforms/copilot.prompt.md.

Run Chat: Install Plugin From Source from the Command Palette. Enter https://github.com/TheQtCompanyRnD/agent-skills (this repo) to clone and install

Install as an extension:

gemini extensions install https://github.com/TheQtCompanyRnD/agent-skills

Or import skills into your project's context file:

# Add to GEMINI.md (loaded automatically)
echo '@skills/qt-qml-review/SKILL.md' >> GEMINI.md

Neither tool reads skill directories. Copy the skill in as a rule:

# Cursor: the full SKILL.md becomes a project rule
mkdir -p .cursor/rules/qt-qml-review
cp skills/qt-qml-review/SKILL.md .cursor/rules/qt-qml-review/RULE.md

# Windsurf: use the condensed variant from platforms/
mkdir -p .windsurf/rules
cp skills/qt-qml-review/platforms/windsurf.md .windsurf/rules/qt-qml-review.md

Windsurf variants exist for qt-cpp-review, qt-qml-review, qt-qml-docs and qt-qml. Scripts and subagents don't run on these tools, so only the static guidance applies.

A browsable documentation site is built from the docs/ directory using MkDocs Material and published at:

https://doc.qt.io/agentictools/

The canonical repo is on Gerrit (codereview.qt-project.org/qtai/qtaiskills); doc.qt.io builds are driven by the qtaiskills Rundeck job in tqtc-doctools on each push to the publishing branch.

Requires Python 3.12+:

pip install -r requirements-docs.txt
mkdocs serve

Then open http://127.0.0.1:8000/agent-skills/ — the dev server live-reloads on every save.

mkdocs build --strict

Output goes to site/. --strict fails on broken links and malformed nav, matching the GitHub Actions check.

Skill pages are generated at build time by gen_skills.py: each skills/<skill-name>/SKILL.md becomes a page, a row in the skills overview table and an entry in the Skills nav. Adding the skill directory is enough; don't create docs/skills/<skill-name>.md or edit nav: in mkdocs.yml. Run mkdocs serve to verify.

See CONTRIBUTING.md for guidelines on authoring new skills, including:

• Naming conventions and frontmatter requirements
• How to structure progressive disclosure
• Testing skills across AI coding tools
• When and how to create platform-specific variants

BSD-3-Clause — see LICENSE for details.