feat(codemod-html-defaults): codemod removing redundant native HTML attribute defaults

[smndhm]

New package @dume/codemod-html-defaults: a CLI + library that removes native HTML attributes whose literal value is already the WHATWG spec default — e.g. <input type="text"> → <input>, <form method="get"> → <form>, <script type="text/javascript"> → <script>.

Usage

npx codemod-html-defaults src/                    # dry-run (default): per-attribute report, writes nothing
npx codemod-html-defaults --write src/            # apply
npx codemod-html-defaults --ext html,njk --write . # opt template files in when walking directories

Runs are idempotent, and edits are byte-preserving outside the removed attribute ranges (parse5 + source locations), so HTML-shaped templates like the Nunjucks files in apps/articles are safe.

Where the defaults come from (nothing from memory)

• Primary source: html-enumerated-attributes (MIT, unified collective), which encodes the WHATWG missing value defaults as state equivalence classes. A value is removable only when it belongs to the same state group as the missing value default.
• A small spec-linked supplementary table in src/defaults.ts for the non-enumerated cases (script[type], style[type], link[type]), cross-checked with html-minifier-terser's redundant-attribute handling.
• Empirical verification in tests against jsdom's WHATWG reflection (input.type === 'text', form.method === 'get', button.type === 'submit', …).

Conditional defaults — never applied flatly

• button[type="submit"]: never removed in v1, only reported (behavior depends on form association; explicit types are often kept for readability).
• target="_self" on a/area/form: removed only in a full document with no <base target>; fragments are always skipped (a partial may be included under a layout with <base>); target on <base> itself is never touched.
• link[type="text/css"]: removed only when the element's static rel contains the stylesheet token.

Never touched

Dynamic values ({{ }}, {% %}, <%= %>, ${}) and binding attribute names (:type, v-bind:, ng-*), boolean attributes and empty values, custom elements, foreign (SVG/MathML) content, and start tags containing duplicate attributes.

Verification

• 73 tests, 96.7% statement coverage: removal matrix, dynamic/boolean guards, conditional cases (button in/out of form written before any rule), normalization (case, quotes, no trimming), idempotence, duplicate-attribute safety, CLI dry-run/write.
• Functional no-op proven on form submission behavior (method/enctype/type/required/elements identical before and after in jsdom) — not just snapshots.
• Dry-run on this repo's own apps/: 7 files scanned, 0 changes (repo already clean; type="module" correctly untouched).
• pnpm lint, manypkg check and the package's vite build all pass. Note: slides#build fails in turbo on main too (marp-cli browser discovery, environment issue) — unrelated to this change.

A changeset is included for the initial 1.0.0 release.

🤖 Generated with Claude Code

https://claude.ai/code/session_01QjF61pCbakF8bMW4XaaFeW

---

Generated by Claude Code

[changeset-bot]

🦋 Changeset detected

Latest commit: 7893919

The changes in this PR will be included in the next version bump.

This PR includes changesets to release 1 package

Name	Type
@dume/codemod-html-defaults	Major

Not sure what this means? Click here to learn what changesets are.

Click here if you're a maintainer who wants to add another changeset to this PR