wpt-evaluator-conformance

name	wpt-evaluator-conformance
description	Evaluate a WPT test file against extracted normative requirements from its governing spec. Use when a spec URL is available and conformance to the spec needs to be judged. Pairs with the wpt-evaluator skill, which checks against WPT documentation guidance rather than spec conformance.

WPT Evaluator — Spec Conformance

This skill evaluates a single WPT test file against a list of normative requirements that have already been extracted from the test's governing specification. It is an advisory signal for human reviewers, not a merge gate.

The requirements list is provided to you as XML in the prompt. Each <requirement> carries an id (e.g. R3), a category, and the normative statement itself. Your job is to read the test file and judge whether the test's assertions align with each requirement.

When to use

• A spec URL was provided for the test under evaluation, and a requirements XML has already been extracted from it.
• Use alongside the wpt-evaluator skill (the docs/style pass), not in place of it.

Do not use this skill as a substitute for human review, running the test, or running wpt lint.

Inputs

• The path to a single WPT test file.
• A <requirements_list> XML block in the prompt, containing one or more <requirement id="..."> elements.

Outputs

For each finding:

• Severity:
  • error — the test contains an assertion that contradicts a normative requirement (the test asserts behavior the spec explicitly forbids, or asserts the opposite of what the spec requires).
  • warn — the test contains an assertion about behavior that is not normatively specified in the provided requirements. There may be legitimate reasons for such tests, but the absence of a matching requirement is worth surfacing.
  • info and nit are not used by this skill.
• Line reference into the test file under evaluation.
• Short evidence quote from the test file (the assertion in question).
• Source citation: the requirement ID (e.g. R3) that the finding relates to, formatted as requirements.xml#R3. For warn findings where no requirement matches, use requirements.xml#none-matched.
• One-sentence summary of the contradiction or gap.

Prohibited outputs

The same hard prohibitions apply as in the wpt-evaluator skill:

1. No composite score. Do not aggregate findings into a single number, grade, or pass/fail verdict.
2. No proposed fixes. Findings describe what is wrong and why, never how to fix it. No "should use X instead", no suggested rewrites, no concrete code blocks that are not already in the test file.
3. No invented requirement IDs. Cite only IDs that appear in the provided <requirements_list>, or use requirements.xml#none-matched for warn findings.

If you find yourself writing "should be X" or attaching a code suggestion, stop. State the problem and cite the requirement; the human reviewer decides the remediation.

Procedure

1. Read the test file at the provided path.
2. Identify the test's assertions. For a testharness test these are typically assert_* calls inside test() / promise_test() blocks. For a reftest, the assertion is the visual comparison between test and reference. For an idl test, the assertions are produced by IdlArray.test(). Note the line numbers.
3. For each assertion, judge whether it aligns with the <requirements_list>:
   • Contradicts a requirement → emit an error finding, citing the requirement ID.
   • Matches a requirement → no finding (this is the expected case).
   • Does not correspond to any requirement in the list → emit a warn finding, citing requirements.xml#none-matched.
4. For each requirement in the list that is not exercised by any assertion, do not emit a finding. Coverage gaps are out of scope for this skill — we are judging the test as-written, not what is missing.
5. Submit findings by calling report_conformance_complete with findings and input_scope payloads. Each finding must populate the fields described in the Outputs section above. Before submitting, verify no finding violates the prohibited outputs.

Tracking input scope

Record every file you opened in service of the evaluation, along with its byte size (wc -c <path>). Include the test file itself. Do NOT count files only used for navigation.

For each tracked file, classify it as one of:

• skill — this SKILL.md.
• test — the test file under evaluation.
• requirements — the requirements XML (passed in the prompt, not read from disk; report its byte size as the byte length of the XML string).
• dependency — a file referenced by the test and read on demand.

You submit this data as part of the input_scope payload to the report_conformance_complete tool. The wpt-gen CLI renders the final Input scope table; do NOT format any of this yourself.