wpt-generator

name	wpt-generator
description	Generate Web Platform Tests (WPT) from minimal XML test suggestions. The agent will autonomously determine the test type and implementation details by analyzing existing repository paradigms. Use when the user asks to generate a Web Platform Test based on a test suggestion.

Web Platform Test Generator

This skill enables an ADK agent to generate Web Platform Tests (WPT) from minimal XML test suggestions. Because the test suggestion only contains high-level requirements, you must rely on existing codebase paradigms to determine how the test should be written.

Workflow

Temporary Files Policy: If you need to create any temporary files during research, prototyping, or debugging, use the write_file tool to place them exclusively inside a .wpt-generator-tmp/ directory at the root of the repository. Do not scatter temporary files across the codebase.

When asked to generate a WPT from an XML test suggestion, follow these steps:

1. Parse the test suggestion

Extract the following elements from the <test_suggestion> XML snippet provided by the user: {% if has_web_feature_id %}

• <web_feature_id>: Used to find where the test should live. {% endif %}
• <description>: The underlying requirement or specification behavior to test.
• <spec_url> (can be multiple): A link to the specification. You MUST include these exact URLs in the generated test (using <link rel="help" href="..."> for HTML tests, or as a single-line comment for .js tests).

2. Locate the Test Directory

{% if has_web_feature_id %} Determine where this test belongs in the repository by finding the corresponding WEB_FEATURES.yml file.

1. Use the search_feature_tests tool with the <web_feature_id> to find existing tests for this feature.
2. Review the output to determine the target directory. {% else %} Determine where this test belongs in the repository based on the specification URL and the feature description.
3. Broadly search the repository for keywords related to the API or feature.
4. Review the findings to determine the most logical target directory. {% endif %}

3. Research Existing Paradigms & Determine Test Type

Since you are not provided with explicit steps or a test type, you MUST research how similar tests are written for this feature, both in the target directory and across the broader codebase. {% if has_web_feature_id %}Avoid "Tunnel Vision": Do not restrict your research solely to the output of search_feature_tests. {% endif %}1. Use the list_directory and search_files tools to explore existing tests in the target directory. 2. Broaden your search: Search the broader repository (or related parent directories) for the API name or feature (e.g., fetchLater) to find existing test ecosystems, helper files (resources/), or data-driven testing paradigms that might live in adjacent directories. 3. IDL Check: If the <description> involves testing interface exposure, attributes, or methods, you MUST check the repository's interfaces/ directory for a corresponding .idl file (e.g., interfaces/crash-reporting.idl). If it exists, this strongly indicates you should use idlharness.js instead of manual boolean assertions. 4. Read 1 or 2 existing tests that seem related to the <description>. Treat these as "Golden Examples". 5. Based on the requirement and the golden examples, decide on the best Test Type:

• Testharness test: Best for JS APIs, parsing, DOM manipulation, or computed CSS values.
• Reftest: Best for visual/rendering layout matching.
• Crashtest: Best for ensuring no browser crash occurs.
• wdspec test: Best for verifying WebDriver Classic or WebDriver BiDi protocols. Written in Python using pytest.
• Manual test (ABSOLUTE LAST RESORT): You MUST NOT write a manual test unless it is fundamentally impossible to test the behavior using current automated tooling (e.g., standard JS, testharness.js, or testdriver.js). This format is strictly reserved for behaviors that require unavoidable physical human interaction or OS-level interventions (e.g., forcefully crashing a browser process via the OS task manager). If automation is possible, you MUST automate it.

4. Determine the Appropriate File(s) & Name

Before creating a new file, rigorously check if the test logic belongs in existing files. Minimize boilerplate by reusing existing files whenever possible.

1. Analyze Directory Paradigms: Check if the target directory splits tests by category (e.g., separating valid vs. invalid values, or computed vs. parsing behavior).
2. Split the test suggestion if Necessary: A single XML test suggestion might encompass multiple test categories. If the directory separates testing into distinct files (e.g., feature-valid.html and feature-invalid.html), you MUST split the test logic across the respective existing files rather than creating a single, monolithic new file.
3. Reftest Reference Search: If you selected Reftest, you MUST search the target directory (and any reference/ subdirectories) for existing reusable reference files (e.g., ref-filled-green-100px-square.xht) before deciding to create a new one. Do NOT generate a duplicate reference file if a suitable one exists.
4. Create New Only When Necessary: Only if no logical match is found (even after considering splitting and manual consolidation), plan to create a new file. Consult references/wpt_style_guide.md to determine the correct filename extension and suffixes (e.g., .html, .window.js, .any.js) based on your chosen test type. Name the file logically based on the {% if has_web_feature_id %}<web_feature_id>{% else %}feature description{% endif %}.
5. File Existence Check: Before using write_file to create a new test file, you MUST verify that the proposed filename does not already exist in the target directory (e.g., by using the list_directory tool).
6. Naming Conflicts: If a file with the proposed name already exists, you MUST either append the new test logic to the existing file (if logically appropriate) or generate a new unique filename by incrementing a numerical suffix (e.g., changing -001.html to -002.html) to prevent overwriting existing work.
7. Existing Test Expansion: If you detect that the core assertion of the WPT test suggestion is already present in an existing test (i.e. a test perfectly matching the test suggestion exists), you MUST ensure all specification edge cases, permutations, or multi-level DOM configurations are thoroughly exercised, and expand the existing test file as necessary instead of creating a redundant new test.
   • Handling .tentative files: If an existing .tentative file matches your test suggestion, append your new test cases directly to it. Do NOT remove the .tentative suffix or rename the file to "upgrade" it unless explicitly instructed to do so by the user.

5. Load References & Generate the Test

Before writing any code, you MUST read the appropriate style guides to ensure correct formatting and syntax:

• For general guidelines (apply to all): See wpt_style_guide.md
• If Testharness: See testharness_style_guide.md
• If Reftest: See reftest_style_guide.md
• If Crashtest: See crashtest_style_guide.md
• If the test requires simulated user interaction (clicks, typing, gestures) or tests hardware/device APIs (like Web MIDI, Web Bluetooth): See automation_guide.md
• If the test is a wdspec test (testing the WebDriver protocol itself): See wdspec_guide.md
• If the test strictly requires a human operator and cannot be automated: See manual_test_style_guide.md
• If the test involves Web IDL interfaces (e.g., testing [Exposed] attributes, method existence, or interface exposure): See idlharness_guide.md
• If the test requires cross-origin requests, custom HTTP headers, specific status codes, or dynamic server logic: See server_features_guide.md

Write the appropriate WPT test to strictly satisfy the <description> using the write_file tool:

• CRITICAL RULE: Style Guides > Golden Examples: Existing tests ("Golden Examples") sometimes contain legacy code and violate current best practices. You should prioritize the explicit rules in the style guides over the paradigms found in surrounding files. Use existing files to understand the domain logic, but rely exclusively on the style guides for the implementation syntax.
• Omit HTML Boilerplate: Unless the test strictly requires attaching attributes to the root elements, you MUST omit standard <html>, <head>, and <body> tags in your generated .html files (including references) to keep tests focused and concise, even if "Golden Examples" include them. Start directly with <!DOCTYPE html> and <meta charset="utf-8">.
• Deduce Expectations: Carefully deduce the exact pass/fail condition and assertions from the <description>. If the requirement is highly complex or vague, use the fetch_spec_content tool to fetch the <spec_url> text to gain deeper context before writing the test.
• CRITICAL RULE: Minimize Specification Boilerplate: Only use the APIs strictly required to trigger the behavior described in the <description>. Do not include optional features or initialization boilerplate from the spec unless explicitly required by the core assertion.
• CRITICAL RULE: Domain Helpers > Golden Examples: Check if a built-in helper exists in the local resources/ directory to avoid repetitive boilerplate. Even if your "Golden Example" writes out boilerplate logic manually (e.g., manually polling, fetching, resolving a sequence of events, or establishing positive controls), you MUST aggressively replace that boilerplate if a higher-level abstraction exists in a local helper file. When you identify a helper file, you MUST perform an exhaustive audit of its entire exported API (e.g., reading the whole helper.js file using read_file) to discover all available utilities (like wait(), delay(), or custom assertions). Do not restrict your usage only to the specific functions the Golden Example used.
  • If testing CSS property parsing, descriptor parsing (e.g., @font-face descriptors), at-rules, inheritance, computed values, or shorthands: See css_testcommon.md
  • If testing CSS Fonts, typography, or synthetic glyph metrics: See css_fonts.md
  • If testing CSS property animatability, interpolation, or discrete flips: See css_animations.md
  • If testing cross-origin network or fetch behaviors via Javascript: See get_host_info.md
  • If testing features controlled by Permissions Policy (formerly Feature Policy): See permissions_policy.md
• Implementation: Write the test logic, setup, and assertions autonomously. When generating tests for multiple permutations or variations of an API, you should consider writing a data-driven test using arrays and loops, as described in testharness_style_guide.md. However, do not over-engineer simple, isolated features. Note: If the target directory lacks examples of your chosen Test Type, rely entirely on the style guides.

6. Validation & Self-Correction (CRITICAL)

Before completing the task, you MUST validate that the code you generated is syntactically correct, properly formatted, and functions as intended.

1. Linting: Use the run_wpt_lint tool on the file you created/modified.

   • If the linter reports errors (e.g., TRAILING WHITESPACE, INDENT TABS), you MUST use the replace_in_file tool to fix the errors and re-run the linter until it passes cleanly. This saves context tokens compared to rewriting the entire file.

2. Execution: If the run_wpt_test tool is available, use it on the file you created/modified. If the user prompt explicitly instructs you to skip running tests, or if the tool is not in your toolset, you MUST SKIP this execution step.

   • CRITICAL RULE - Manual Tests: If the test you created or modified is a manual test (e.g., ends in -manual.html or requires human intervention), you MUST SKIP this execution step entirely. Rely strictly on the linter for manual tests.
   • Analyze the Output: Read the test runner's output carefully.
   • Self-Correct: If the runner reports a Harness Error, SyntaxError, a timeout, or a failure that indicates a flaw in your test logic (e.g., calling an undefined helper function or making an incorrect assertion), you MUST use the replace_in_file or write_file tools to fix the bug, and re-run the test. Use replace_in_file whenever possible.
   • Repeat this execute-and-fix loop until the test executes successfully without syntax or harness errors. Maximum 3 attempts. If the test still fails after 3 correction attempts, stop debugging and proceed to finalize. (Note: If the test fails because the browser genuinely does not support the feature, that is acceptable—your goal is to ensure the test code itself is valid.)

{% if has_web_feature_id %}

7. Map the Test in WEB_FEATURES.yml

Every generated test file MUST be explicitly mapped to the target <web_feature_id> (from Step 1) in the directory's WEB_FEATURES.yml file.

1. Check for File: Look for WEB_FEATURES.yml in the directory where you created or modified the test using the list_directory tool.
2. Create if Missing: If the file does not exist, create it using the write_file tool with the following structure:

   features:
   - name: <web_feature_id>
     files:
     - <new_test_file_name>
   

3. Update if Existing: If it exists, read it using read_file and update it using write_file to append your new test file name to the files: list under the matching <web_feature_id>. (If the test is already covered by an existing wildcard pattern belonging to the correct feature, you don't need to list it individually).
4. Prevent Collisions (CRITICAL): Carefully review the other web feature IDs defined in the same WEB_FEATURES.yml file. If another feature uses a wildcard (like - "**" or - "*.html") that would accidentally match your newly created test file, you MUST explicitly exclude your test from that feature by adding a negation line (e.g., - "!<new_test_file_name>") to its files: list.

{% endif %}

{% if has_web_feature_id %}8{% else %}7{% endif %}. Finalizing

• Ensure standard WPT scripts are included properly (if applicable) using absolute paths from the root server.
• Ensure crashtests end with -crash.html if creating a new crashtest file.
• Clean Up: You MUST explicitly delete any temporary files you created in .wpt-generator-tmp/ using the delete_file tool to ensure no temporary prototypes, scripts, or intermediate files are left behind in the repository.
• Once the test is validated{% if has_web_feature_id %} and the WEB_FEATURES.yml file is mapped{% endif %}, your task is complete.