Drupal Agent Toolkit
Reusable Copilot agents, skills, and scoped rules for Drupal development and site maintenance. This plugin contains generic Drupal guidance only; project paths, commands, policies, and site details belong in each Drupal repository.
Install the toolkit in VS Code
Important
Start here. Install directly from the Command Palette, not from Browse customizations or GitHub Feed. First-time users: complete the setup checklist before installing.
-
Click View > Command Palette.
-
Type Chat: Install Plugin from Source and press Enter.
-
Paste this URL into the source box and press Enter:
https://github.com/joao-silva-ntt/Drupal-Agent-Toolkit -
If asked where to install it, select User to make it available across your projects.
-
Review the repository and trust prompt. Confirm only if you trust the source, then wait for installation to finish.
Setup checklist
First time installing a plugin? Open this checklist.
Get ready
- Use an up-to-date VS Code with GitHub Copilot Chat, signed in to Copilot.
- Install Git on the computer running VS Code. For Windows with WSL, install Git on Windows too; Git inside WSL alone is not enough for the Windows-side installer.
- After installing Git, close every VS Code window, including any Agents windows, and reopen VS Code.
Enable plugins
- In VS Code, click View > Command Palette.
- Type Preferences: Open Settings (UI) and press Enter.
- Click the Search Settings box at the top and type
chat.plugins.enabled. - Check Chat > Plugins: Enabled.
Then follow Install the toolkit in VS Code above.
Start using it
- Open your Drupal project and start a new Copilot chat.
- Select Copilot as the agent harness if prompted.
- Open the agent picker in the chat input and choose a specialist, such as Backend Developer or Drupal Architect.
- Type
/drupal-agent-toolkit:in the chat input to see the toolkit's skills. - Start with a read-only request, such as: "Explain this project's structure without changing any files."
Installation makes the toolkit available in Copilot; it does not copy files into your project. Complete the project-specific setup below before asking agents to make substantial changes.
Project-specific setup
The plugin does not assume a particular Drupal version, theme, deployment, toolchain, or QA target. For substantial work, provide the project-specific facts and approved commands in:
docs/AGENT_PROJECT.md- human-readable architecture, rules, approvals, and exclusions.docs/agent-project.json- project paths and optional, explicitly reviewed QA configuration.
Start from docs/agent-project.example.json.
If a project profile is absent or incomplete, agents should inspect safely and
ask about material unknowns instead of borrowing assumptions from another
project.
Optional QA runner
Every toolkit agent checks test readiness on first use in a project. It inspects existing accessibility and performance tests without installing anything or running an audit. Missing setup is explained and requires your approval. Read-only/planning agents direct setup to Accessibility Expert or Performance QA.
With permission, readiness evidence is saved in docs/AGENT_PROJECT.md for
other agents to reuse. Relevant file or environment changes trigger a recheck;
a teammate's record does not verify your machine's installed tools. Optional
test setup does not block unrelated work. This is an agent instruction, not
an automatic VS Code startup hook.
The agents handle accessibility and performance test setup for you. Select Accessibility Expert or Performance QA and ask:
Check whether this project's accessibility and performance tests are ready. If setup is missing, explain what you need approval for, then set it up for me and run the checks against my approved site.
The agent checks existing tests first, asks for missing site details and approval, then configures the tools and installs approved dependencies itself. You do not need to copy files or edit test configuration manually. Existing tests are preserved, and site fixes require separate approval.
Installing the plugin does not copy files into the Drupal repository. To use the bundled suites, the agent performs the approved project setup described in the browser QA guide. Nothing is installed or run merely by enabling the plugin. The example profile disables all QA suites; only approved suites are enabled. The runner itself does not install dependencies, start services, or make a site read-only.
Run its isolated standard-library tests from the plugin checkout with:
PYTHONDONTWRITEBYTECODE=1 python3 -m unittest discover \
-s runtime/.github/scripts/tests -v
Toolkit validation
The QA agents include a read-only setup diagnostic, independent suite setup, optional Firefox/WebKit accessibility checks, keyboard/layout checks and repeated-sample performance budgets. Runs preserve separate JSON/HTML reports. See browser QA for details.
Pull requests and pushes run metadata/portability/link checks, QA-runner tests, and isolated browser fixtures after a clean dependency install on Linux and Windows. Browser checks cover Chromium plus optional Firefox/WebKit in CI. They do not audit a project site or establish accessibility conformance.
Safety and scope
- Agents and skills provide guidance; they do not prove compliance or replace project review.
- QA agents are read-only by default and should report untested or blocked coverage explicitly.
- This plugin includes no hooks or MCP servers. The optional QA scripts run only when explicitly invoked.
- This is a community project and is not an official Drupal, GitHub, or Microsoft product.
See docs/AGENT_TOOLKIT.md for the toolkit's
structure and adoption notes.