chore: adopt oxfmt and oxlint with a pre-commit hook

.githooks/pre-commit

@@ -0,0 +1,65 @@
+#!/bin/sh
+# Blocks a commit unless the whole codebase, exactly as it will be committed, is formatted and
+# passes lint.
+#
+# oxfmt is the formatter (.oxfmtrc.json); oxlint is the linter (.oxlintrc.json). Wired up by
+# the "prepare" script in package.json (git config core.hooksPath .githooks) on npm install.
+# Emergency bypass: git commit --no-verify
+#
+# This file is shared verbatim by jetstreamapp/sf-formula-parser, jetstreamapp/soql-parser-js,
+# jetstreamapp/simple-excel and jetstreamapp/simple-xml - keep every copy in sync when changing it.
+
+root=$(git rev-parse --show-toplevel) || exit 1
+OXFMT="$root/node_modules/.bin/oxfmt"
+OXLINT="$root/node_modules/.bin/oxlint"
+
+for bin in "$OXFMT" "$OXLINT"; do
+  if [ ! -x "$bin" ]; then
+    echo "pre-commit: $bin not found - run 'npm install' first." >&2
+    exit 1
+  fi
+done
+
+# The checks run against a snapshot of the index rather than the working tree. The index is
+# exactly what the commit will contain, so a file formatted on disk but never re-staged cannot
+# slip through, and unstaged edits cannot block a clean commit. git points GIT_INDEX_FILE at the
+# right index for `commit -a` and `commit <path>` too, and checkout-index honours it. Every
+# tracked file is exported, so the whole codebase is checked, not only the files being changed.
+snapshot=$(mktemp -d "${TMPDIR:-/tmp}/pre-commit.XXXXXX") || exit 1
+trap 'rm -rf "$snapshot"' EXIT
+trap 'exit 1' HUP INT TERM
+
+git -C "$root" checkout-index --all --prefix="$snapshot/" || exit 1
+
+# Some lint rules resolve imports, so the snapshot shares the installed dependencies.
+ln -s "$root/node_modules" "$snapshot/node_modules" || exit 1
+
+cd "$snapshot" || exit 1
+
+# Both run even when the first fails, so one commit attempt reports everything to fix.
+status=0
+
+if ! "$OXFMT" --check; then
+  cat >&2 <<'MSG'
+
+Commit blocked: the files listed above are not formatted.
+
+  npm run format     # format the repo
+  git add -u         # stage the result
+MSG
+  status=1
+fi
+
+if ! "$OXLINT"; then
+  cat >&2 <<'MSG'
+
+Commit blocked: the files listed above have lint errors.
+
+  npm run lint       # review the full report
+  npm run lint:fix   # apply the fixes oxlint can make automatically
+  git add -u         # stage the result
+MSG
+  status=1
+fi
+
+exit $status

.github/workflows/ci.yml

@@ -22,6 +22,9 @@ jobs:
           cache: npm
 
       - run: npm ci
+      # The pre-commit hook runs these too, but it can be bypassed with --no-verify
+      - run: npm run format:check
+      - run: npm run lint
       - run: npm run typecheck
       # build before test, matching sf-formula-parser and soql-parser-js: tests run against a
       # build that is known to have succeeded.

.prettierrc → .oxfmtrc.json

@@ -1,4 +1,5 @@
 {
+  "$schema": "./node_modules/oxfmt/configuration_schema.json",
   "printWidth": 140,
   "semi": true,
   "tabWidth": 2,
@@ -7,5 +8,6 @@
   "trailingComma": "all",
   "bracketSpacing": true,
   "arrowParens": "avoid",
-  "insertPragma": false
+  "sortPackageJson": false,
+  "ignorePatterns": []
 }

.oxlintrc.json

@@ -0,0 +1,33 @@
+{
+  "$schema": "./node_modules/oxlint/configuration_schema.json",
+  // typescript, unicorn, oxc and import apply everywhere. vitest is enabled only in the test
+  // override below, so its rules cannot fire on source files.
+  "plugins": ["typescript", "unicorn", "oxc", "import"],
+  // correctness, suspicious and perf catch defects, so they are errors. style and pedantic stay
+  // off because oxfmt owns layout and the rest is opinion rather than defect detection.
+  // restriction and nursery are off by default, listed so the whole policy reads in one place.
+  "categories": {
+    "correctness": "error",
+    "suspicious": "error",
+    "perf": "error",
+    "pedantic": "off",
+    "style": "off",
+    "restriction": "off",
+    "nursery": "off"
+  },
+  "overrides": [
+    {
+      // The release script polls external state; each iteration must observe the result of the
+      // previous one, so the sequential awaits are deliberate.
+      "files": ["scripts/release.mjs"],
+      "rules": {
+        "no-await-in-loop": "off"
+      }
+    },
+    {
+      "files": ["src/__tests__/**"],
+      "plugins": ["vitest"]
+    }
+  ],
+  "ignorePatterns": ["dist/**"]
+}

AGENTS.md

@@ -4,6 +4,46 @@
 
 Use conventional commit style messages.
 
+## Formatting
+
+**ALWAYS run `npm run format` after making any change.** This is not optional and applies to every
+file you touch — source, tests, docs, config, JSON, YAML and Markdown alike.
+
+oxfmt is the formatter; its settings live in `.oxfmtrc.json`. It reads `.gitignore`, so build output
+is skipped automatically.
+
+- `npm run format` — format the repo in place
+- `npm run format:check` — verify without writing
+
+## Linting
+
+**ALWAYS run `npm run lint` after making any change** and fix what it reports.
+
+oxlint is the linter; its settings live in `.oxlintrc.json`. It lints `src/` and `scripts/` — `dist/`
+is excluded via `ignorePatterns`.
+
+- `npm run lint` — report violations
+- `npm run lint:fix` — apply the fixes oxlint can make automatically
+
+The `correctness`, `suspicious` and `perf` categories are errors. `style` and `pedantic` are
+deliberately off: oxfmt already owns layout, and the remainder is opinion rather than defect
+detection.
+
+Prefer fixing the code over silencing the rule. When a rule genuinely misfires, disable it at the
+narrowest scope that works — an `oxlint-disable-next-line` comment or a file-glob entry in
+`overrides` rather than repo-wide — and record why inline. The existing exceptions cover the
+sequential polling loop in `scripts/release.mjs` and the in-place `reverse()` in `src/parse.ts`,
+where the suggested `toReversed()` is above the ES2022 build target.
+
+## Commit hook
+
+A `pre-commit` hook in `.githooks/` runs both the format check and oxlint against the whole codebase
+and **blocks the commit** if either fails, so skipping those steps will stop the commit rather than
+slip through. It checks a snapshot of the index, not the working tree, so it sees exactly what the
+commit will contain: a file formatted on disk but not re-staged still blocks, and unstaged edits
+never do. The hook is wired up by the `prepare` script (`git config core.hooksPath .githooks`) on
+`npm install`. Emergency bypass: `git commit --no-verify`.
+
 ## Releasing
 
 Releases are cut from `main` by CI, kicked off from your terminal:
@@ -48,4 +88,5 @@ hook failing mid-release would abort after the npm publish.
 `changelog.yml` workflows are shared verbatim with
 [sf-formula-parser](https://github.com/jetstreamapp/sf-formula-parser) and
 [soql-parser-js](https://github.com/jetstreamapp/soql-parser-js) — keep the copies in sync when
-changing them.
+changing them. `.githooks/pre-commit` and `.oxfmtrc.json` are shared the same way, with
+[simple-excel](https://github.com/jetstreamapp/simple-excel) as well.

CHANGELOG.md

@@ -4,6 +4,21 @@ All notable changes to this project will be documented in this file.
 
 ## [Unreleased]
 
+### Changed
+
+- **Replaced Prettier with [oxfmt](https://oxc.rs/docs/guide/usage/formatter.html) and added
+  [oxlint](https://oxc.rs/docs/guide/usage/linter) as the project linter**, matching
+  sf-formula-parser, soql-parser-js and simple-excel. Settings live in `.oxfmtrc.json` and
+  `.oxlintrc.json`, with `npm run format`, `format:check`, `lint` and `lint:fix`. oxlint runs the
+  `correctness`, `suspicious` and `perf` categories as errors
+- **The `pre-commit` hook blocks a commit unless the whole codebase, as staged, is formatted and
+  passes lint.** `npm install` wires it up by pointing `core.hooksPath` at `.githooks/`. Emergency
+  bypass: `git commit --no-verify`
+- **CI now gates on `format:check` and `lint`**, since the pre-commit hook can be bypassed
+- Internal cleanups surfaced by oxlint, with no behaviour change: `addChild` moved to module scope
+  so it is no longer recreated on every `parse()` call, and `parseAttributes` dropped a `coerce`
+  parameter left unused since attribute values stopped being coerced in 1.1.0
+
 ## [1.2.0] - 2026-09-30
 
 ### Added

README.md

@@ -51,15 +51,15 @@ const result = parse(soapResponse, {
 
 #### ParseOptions
 
-| Option                | Type      | Default | Description                                              |
-| --------------------- | --------- | ------- | -------------------------------------------------------- |
-| `trimValues`          | `boolean` | `true`  | Trim whitespace from text content                        |
-| `ignoreAttributes`    | `boolean` | `false` | Skip attributes entirely                                 |
-| `removeNSPrefix`      | `boolean` | `false` | Strip namespace prefixes (`soap:Body` → `Body`)          |
-| `attributeNamePrefix` | `string`  | `'@_'`  | Prefix prepended to attribute names in the output object |
-| `parseTagValue`       | `boolean` | `true`  | Coerce numeric/boolean text values to their JS types     |
-| `processEntities`     | `boolean` | `true`  | Decode XML entities (`&amp;` → `&`, `&lt;` → `<`, etc.)  |
-| `maxDepth`            | `number`  | `256`   | Maximum nesting depth allowed. Throws if exceeded        |
+| Option                | Type      | Default | Description                                                          |
+| --------------------- | --------- | ------- | -------------------------------------------------------------------- |
+| `trimValues`          | `boolean` | `true`  | Trim whitespace from text content                                    |
+| `ignoreAttributes`    | `boolean` | `false` | Skip attributes entirely                                             |
+| `removeNSPrefix`      | `boolean` | `false` | Strip namespace prefixes (`soap:Body` → `Body`)                      |
+| `attributeNamePrefix` | `string`  | `'@_'`  | Prefix prepended to attribute names in the output object             |
+| `parseTagValue`       | `boolean` | `true`  | Coerce numeric/boolean text values to their JS types                 |
+| `processEntities`     | `boolean` | `true`  | Decode XML entities (`&amp;` → `&`, `&lt;` → `<`, etc.)              |
+| `maxDepth`            | `number`  | `256`   | Maximum nesting depth allowed. Throws if exceeded                    |
 | `strict`              | `boolean` | `false` | Throw on malformed XML (unclosed, mismatched, or extra closing tags) |
 
 ### `build(obj: Record<string, unknown>, options?: BuildOptions): string`
@@ -81,12 +81,12 @@ const xml = build(
 
 #### BuildOptions
 
-| Option                | Type      | Default | Description                                                |
-| --------------------- | --------- | ------- | ---------------------------------------------------------- |
-| `format`              | `boolean` | `false` | Pretty-print with indentation                              |
-| `ignoreAttributes`    | `boolean` | `false` | Skip attributes when building XML                          |
-| `attributeNamePrefix` | `string`  | `'@_'`  | Prefix used to identify attribute keys in the input object |
-| `indentBy`            | `string`  | `'  '`  | Indentation string used when `format` is `true`            |
+| Option                | Type      | Default | Description                                                    |
+| --------------------- | --------- | ------- | -------------------------------------------------------------- |
+| `format`              | `boolean` | `false` | Pretty-print with indentation                                  |
+| `ignoreAttributes`    | `boolean` | `false` | Skip attributes when building XML                              |
+| `attributeNamePrefix` | `string`  | `'@_'`  | Prefix used to identify attribute keys in the input object     |
+| `indentBy`            | `string`  | `'  '`  | Indentation string used when `format` is `true`                |
 | `declaration`         | `boolean` | `false` | Prepend `<?xml version="1.0" encoding="UTF-8"?>` and a newline |
 
 ## Output Shape