feat(#2975): adopt changeset-fragment workflow to eliminate CHANGELOG conflicts (#2978)

* feat(#2975): adopt changeset-fragment workflow to eliminate CHANGELOG conflicts

Two PRs that both edit `### Fixed` in CHANGELOG.md always conflict on merge.
Recently bit on #2960/#2972 in the same session — fix-the-conflict-and-rebase
tax. Replace the shared-file model with per-PR fragment files that never
share lines.

Implementation built TDD per #2975, vertical slices with structured-IR
assertions throughout:

  scripts/changeset/parse.cjs       - fragment text → typed record + frozen
                                      FRAGMENT_ERROR enum (8 tests)
  scripts/changeset/render.cjs      - fragments → structured IR with
                                      Keep-a-Changelog section ordering
                                      (2 tests)
  scripts/changeset/serialize.cjs   - IR ↔ markdown round-trip pair
                                      (parse(serialize(ir)) === ir,
                                      3 tests)
  scripts/changeset/cli.cjs         - file-I/O wrapper with --json mode;
                                      reads .changeset/, folds into
                                      CHANGELOG.md, deletes consumed
                                      fragments. Idempotent. (1 test)
  scripts/changeset/lint.cjs        - pure verdict (changedFiles, labels)
                                      → { ok, reason } via LINT_REASON
                                      enum. Honors `no-changelog` label.
                                      (5 tests)
  scripts/changeset/new.cjs         - fragment scaffolder with random
                                      adjective-noun-noun filename. Tests
                                      assert via parseFragment round-trip.
                                      (3 tests)

Total: 22 tests, all assertions on typed structured fields. No regex on
text, no String#includes on file content. Lint clean across 356 test files.

Supporting:

  .changeset/README.md              - format spec + workflow docs
  .changeset/eager-hawks-rally.md   - dogfood fragment for THIS PR (will
                                      be the first thing the new release
                                      tool consumes)
  .github/workflows/changeset-required.yml
                                    - CI: every PR runs lint.cjs
  package.json                      - npm run changeset, changelog:render,
                                      lint:changeset
  CONTRIBUTING.md                   - new "CHANGELOG Entries — Drop a
                                      Fragment" section between PR
                                      Guidelines and Testing Standards

Closes #2975

* fix(#2975): address CodeRabbit findings on changeset workflow

7 valid findings (4 Major, 3 Minor); all addressed:

scripts/changeset/parse.cjs
  - Preserve fragment body verbatim. Previously body.trim() ate
    intentional leading whitespace (code blocks, etc.); now trim() is
    used only for the emptiness check, and a single trailing newline
    is stripped (the editor-added one) so well-formed fragments
    round-trip byte-for-byte. Added a regression test asserting a
    code-block-leading body is preserved.

scripts/changeset/cli.cjs
  - Validate flag values during argument parsing. parseArgs now returns
    { ok, opts | error }; rejects `--repo` etc. with no following value
    or with another flag as the value. main() surfaces the error
    message before exiting 2.
  - Handle post-write fragment-deletion failures. After CHANGELOG.md
    is written, any unlink failure is captured into a structured
    deleteFailures list with reason 'fail_fragment_delete'; cmdRender
    returns exitCode=1 with the partial-failure detail instead of
    leaving the changelog updated and fragments behind (which would
    cause double-consumption on rerun).

scripts/changeset/lint.cjs
  - Treat CHANGELOG.md as a linted user-facing path. Direct edits to
    CHANGELOG.md (the bypass route around the new workflow) now fail
    the lint with FAIL_MISSING_FRAGMENT. Added a regression test for
    that case.
  - Use cp.execFileSync instead of cp.execSync for the git diff call.
    Eliminates the shell-interpolation surface on GITHUB_BASE_REF;
    git's own arg parser remains the validator.

scripts/changeset/new.cjs
  - Atomic fragment creation. existsSync() + writeFileSync was racy
    under concurrent invocations. Now writeFileSync uses { flag: 'wx' }
    which fails EEXIST on collision; the random-name retry loop
    catches EEXIST and re-rolls. Throws explicitly after 16 attempts
    rather than silently overwriting.

.changeset/README.md
  - Add language tag `md` to the format example fence (markdownlint
    MD040).

All 25 changeset tests pass; lint clean (356 test files, 0 violations).

* fix(#2975): sanitize --type and validate flag values in new.cjs (CR fixes)

Two CR findings on scripts/changeset/new.cjs:

1. (Minor) `type` was embedded in frontmatter without sanitization. A
   newline in the value (e.g. `--type 'Fixed\ntype: Added'`) would
   corrupt the fragment. scaffoldFragment now validates `type` against
   the Keep-a-Changelog ALLOWED_TYPES set BEFORE writing — same set
   parse.cjs uses on consume. Throws with a typed error referencing
   the allowed values; tests cover the newline case + 4 other
   non-allowed values.

2. (Minor) `--repo` (and other value-taking flags) without a value
   silently set opts.repo to undefined, which produced a cryptic
   ERR_INVALID_ARG_TYPE deep inside path.join. parseArgs now mirrors
   the cli.cjs convention: returns { ok, opts | error }, validates
   that the next token exists and is not itself another flag, and
   surfaces a precise "missing value for --repo" message before exit.
   Added 3 tests: missing-trailing-value, flag-as-value, well-formed.

29 tests pass across the changeset suite (4 new regression tests).

scripts/changeset/cli.cjs

@@ -0,0 +1,191 @@
+#!/usr/bin/env node
+'use strict';
+
+/**
+ * CLI wrapper for the changeset-fragment workflow (#2975).
+ *
+ * Subcommands:
+ *   render --repo <dir> --version V --date D [--json]   Fold .changeset/*.md
+ *                                                       into CHANGELOG.md;
+ *                                                       delete consumed fragments.
+ *
+ * `--json` emits a structured report on stdout — the only contract tests
+ * assert against. Per CONTRIBUTING.md "Prohibited: Raw Text Matching on
+ * Test Outputs", the human formatter is operator-only.
+ */
+
+const fs = require('node:fs');
+const path = require('node:path');
+
+const { parseFragment, FRAGMENT_ERROR } = require('./parse.cjs');
+const { renderChangelog } = require('./render.cjs');
+const { serializeChangelog } = require('./serialize.cjs');
+
+function parseArgs(argv) {
+  const opts = { cmd: null, repo: process.cwd(), version: null, date: null, json: false };
+  if (argv.length === 0) return { ok: true, opts };
+  opts.cmd = argv[0];
+
+  // Pull a value for a value-taking flag, validating that the next token
+  // exists and is not itself another flag (which is the silently-misparsed
+  // case CR called out: e.g. `--repo --json` would consume `--json` as the
+  // repo path).
+  const requireValue = (flag, i) => {
+    const v = argv[i + 1];
+    if (v === undefined || v.startsWith('--')) {
+      return { ok: false, error: `missing value for ${flag}` };
+    }
+    return { ok: true, value: v };
+  };
+
+  for (let i = 1; i < argv.length; i++) {
+    const a = argv[i];
+    if (a === '--json') { opts.json = true; continue; }
+    if (a === '--repo' || a === '--version' || a === '--date') {
+      const r = requireValue(a, i);
+      if (!r.ok) return { ok: false, error: r.error };
+      if (a === '--repo') opts.repo = r.value;
+      else if (a === '--version') opts.version = r.value;
+      else if (a === '--date') opts.date = r.value;
+      i++;
+      continue;
+    }
+    return { ok: false, error: `unknown argument: ${a}` };
+  }
+  return { ok: true, opts };
+}
+
+function listFragmentFiles(changesetDir) {
+  if (!fs.existsSync(changesetDir)) return [];
+  return fs.readdirSync(changesetDir)
+    .filter((f) => f.endsWith('.md') && f !== 'README.md')
+    .map((f) => path.join(changesetDir, f));
+}
+
+function splitChangelog(text) {
+  // Split off the top-level "# Changelog" heading + lead matter (everything
+  // before the first "## [version]" block) from the rest. The rest is the
+  // priorChangelog passed into renderChangelog. The "## [Unreleased]" block,
+  // if present, is dropped (the new release replaces it).
+  const lines = text.split(/\r?\n/);
+  const firstReleaseIdx = lines.findIndex((l) => /^##\s+\[/.test(l));
+  if (firstReleaseIdx === -1) {
+    return { lead: text.replace(/\s+$/, ''), prior: '' };
+  }
+  const lead = lines.slice(0, firstReleaseIdx).join('\n').replace(/\s+$/, '');
+  let priorStart = firstReleaseIdx;
+  // Skip the [Unreleased] block if present — it's a placeholder, not a release.
+  if (/^##\s+\[Unreleased\]/i.test(lines[firstReleaseIdx])) {
+    let j = firstReleaseIdx + 1;
+    while (j < lines.length && !/^##\s+\[/.test(lines[j])) j++;
+    priorStart = j;
+  }
+  const prior = lines.slice(priorStart).join('\n').trimStart();
+  return { lead, prior };
+}
+
+function cmdRender(opts) {
+  const repo = path.resolve(opts.repo);
+  const changesetDir = path.join(repo, '.changeset');
+  const changelogPath = path.join(repo, 'CHANGELOG.md');
+  const fragmentFiles = listFragmentFiles(changesetDir);
+
+  const fragments = [];
+  const failures = [];
+  for (const file of fragmentFiles) {
+    const src = fs.readFileSync(file, 'utf8');
+    const r = parseFragment(src);
+    if (r.ok) fragments.push({ ...r.fragment, file });
+    else failures.push({ file: path.relative(repo, file), reason: r.reason, detail: r.detail || null });
+  }
+
+  if (failures.length > 0) {
+    return { exitCode: 1, report: { consumed: 0, failures } };
+  }
+  if (fragments.length === 0) {
+    return { exitCode: 0, report: { consumed: 0, failures: [] } };
+  }
+
+  const priorText = fs.existsSync(changelogPath) ? fs.readFileSync(changelogPath, 'utf8') : '';
+  const { lead, prior } = splitChangelog(priorText);
+
+  const ir = renderChangelog({
+    fragments,
+    version: opts.version,
+    date: opts.date,
+    priorChangelog: prior || null,
+  });
+  const releaseBlock = serializeChangelog(ir);
+  const out = [
+    lead || '# Changelog',
+    '',
+    '## [Unreleased]',
+    '',
+    releaseBlock.replace(/\s+$/, ''),
+    '',
+  ].join('\n');
+
+  fs.writeFileSync(changelogPath, out);
+
+  // Delete consumed fragments. If any unlink fails the changelog is written
+  // but the fragment is still on disk, so a re-run would double-consume it.
+  // Surface the partial-failure as exitCode=1 with structured detail so the
+  // operator can manually clean up before retrying.
+  const deleteFailures = [];
+  for (const f of fragments) {
+    try {
+      fs.unlinkSync(f.file);
+    } catch (e) {
+      deleteFailures.push({
+        file: path.relative(repo, f.file),
+        reason: 'fail_fragment_delete',
+        detail: e.code || e.message,
+      });
+    }
+  }
+
+  return {
+    exitCode: deleteFailures.length > 0 ? 1 : 0,
+    report: {
+      consumed: fragments.length - deleteFailures.length,
+      failures: deleteFailures,
+      release: { version: opts.version, date: opts.date },
+    },
+  };
+}
+
+function main() {
+  const parsed = parseArgs(process.argv.slice(2));
+  if (!parsed.ok) {
+    process.stderr.write(`${parsed.error}\n`);
+    process.stderr.write('usage: changeset/cli.cjs render --repo <dir> --version V --date D [--json]\n');
+    process.exit(2);
+  }
+  const { opts } = parsed;
+  if (opts.cmd !== 'render') {
+    process.stderr.write('usage: changeset/cli.cjs render --repo <dir> --version V --date D [--json]\n');
+    process.exit(2);
+  }
+  if (!opts.version || !opts.date) {
+    process.stderr.write('--version and --date are required for render\n');
+    process.exit(2);
+  }
+
+  const { exitCode, report } = cmdRender(opts);
+  if (opts.json) {
+    process.stdout.write(JSON.stringify(report, null, 2) + '\n');
+  } else {
+    process.stdout.write(`Consumed: ${report.consumed} fragment(s)\n`);
+    if (report.failures.length > 0) {
+      process.stdout.write(`Failures: ${report.failures.length}\n`);
+      for (const f of report.failures) {
+        process.stdout.write(`  ${f.file}: ${f.reason}${f.detail ? ` (${f.detail})` : ''}\n`);
+      }
+    }
+  }
+  process.exit(exitCode);
+}
+
+if (require.main === module) main();
+
+module.exports = { cmdRender, parseArgs, splitChangelog, listFragmentFiles };

scripts/changeset/lint.cjs

@@ -0,0 +1,110 @@
+#!/usr/bin/env node
+'use strict';
+
+/**
+ * Changeset-fragment lint (#2975).
+ *
+ * Pure verdict function evaluateLint({ changedFiles, labels }) returns
+ * { ok, reason } using the LINT_REASON enum. The CLI wrapper calls it with
+ * the PR diff (via `git diff --name-only origin/main...HEAD` or the GitHub
+ * Actions event payload) and the labels list (via the GitHub event).
+ *
+ * Tests assert on the typed verdict, never on free text.
+ */
+
+const LINT_REASON = Object.freeze({
+  OK_FRAGMENT_PRESENT: 'ok_fragment_present',
+  OK_OPT_OUT_LABEL: 'ok_opt_out_label',
+  OK_NO_USER_FACING_CHANGES: 'ok_no_user_facing_changes',
+  FAIL_MISSING_FRAGMENT: 'fail_missing_fragment',
+});
+
+const OPT_OUT_LABEL = 'no-changelog';
+
+// Files counted as "user-facing" — touching any of these requires either a
+// fragment or an explicit opt-out label. Test/CI/docs/lock files do not.
+const USER_FACING_PREFIXES = [
+  'bin/',
+  'get-shit-done/',
+  'agents/',
+  'commands/',
+  'hooks/',
+  'sdk/src/',
+  'sdk/prompts/',
+];
+
+// Exact-match user-facing files. Any direct edit to one of these without a
+// fragment also fails the lint — closes the bypass where a contributor edits
+// CHANGELOG.md directly to sneak past the new workflow.
+const USER_FACING_FILES = new Set(['CHANGELOG.md']);
+
+function isUserFacing(file) {
+  if (USER_FACING_FILES.has(file)) return true;
+  return USER_FACING_PREFIXES.some((p) => file.startsWith(p));
+}
+
+function isFragment(file) {
+  return /^\.changeset\/[^/]+\.md$/.test(file) && !file.endsWith('/README.md');
+}
+
+function evaluateLint({ changedFiles, labels }) {
+  if (changedFiles.some(isFragment)) {
+    return { ok: true, reason: LINT_REASON.OK_FRAGMENT_PRESENT };
+  }
+  if (labels.includes(OPT_OUT_LABEL)) {
+    return { ok: true, reason: LINT_REASON.OK_OPT_OUT_LABEL };
+  }
+  if (!changedFiles.some(isUserFacing)) {
+    return { ok: true, reason: LINT_REASON.OK_NO_USER_FACING_CHANGES };
+  }
+  return { ok: false, reason: LINT_REASON.FAIL_MISSING_FRAGMENT };
+}
+
+function main() {
+  const fs = require('node:fs');
+  const cp = require('node:child_process');
+  // GitHub Actions event payload path
+  const eventPath = process.env.GITHUB_EVENT_PATH;
+  let labels = [];
+  if (eventPath && fs.existsSync(eventPath)) {
+    try {
+      const event = JSON.parse(fs.readFileSync(eventPath, 'utf8'));
+      labels = (event.pull_request?.labels || []).map((l) => l.name);
+    } catch { /* fall through */ }
+  }
+  const base = process.env.GITHUB_BASE_REF || 'main';
+  let changedFiles = [];
+  try {
+    // Use execFileSync with an argv array — the base ref is interpolated
+    // into a refspec argument, but execFileSync does not invoke a shell, so
+    // even a malicious GITHUB_BASE_REF cannot inject shell syntax. The
+    // refspec-bound metacharacters that git itself rejects (e.g. spaces in
+    // ref names) are caught by git's own arg parser.
+    const out = cp.execFileSync(
+      'git',
+      ['diff', '--name-only', `origin/${base}...HEAD`],
+      { encoding: 'utf8' },
+    );
+    changedFiles = out.split('\n').filter(Boolean);
+  } catch (e) {
+    process.stderr.write(`could not compute diff: ${e.message}\n`);
+    process.exit(2);
+  }
+
+  const verdict = evaluateLint({ changedFiles, labels });
+  if (process.argv.includes('--json')) {
+    process.stdout.write(JSON.stringify({ ...verdict, changedFiles, labels }, null, 2) + '\n');
+  } else if (verdict.ok) {
+    process.stdout.write(`ok changeset-lint: ${verdict.reason}\n`);
+  } else {
+    process.stderr.write(`\nERROR changeset-lint: ${verdict.reason}\n`);
+    process.stderr.write(`PR touches user-facing files but does not include a .changeset/*.md fragment.\n`);
+    process.stderr.write(`Run \`npm run changeset\` to create one, or add the \`${OPT_OUT_LABEL}\` label\n`);
+    process.stderr.write(`if this PR genuinely has no user-facing impact (test refactor, CI tweak, etc.).\n`);
+  }
+  process.exit(verdict.ok ? 0 : 1);
+}
+
+if (require.main === module) main();
+
+module.exports = { evaluateLint, LINT_REASON, OPT_OUT_LABEL, isUserFacing, isFragment };

scripts/changeset/new.cjs

@@ -0,0 +1,137 @@
+#!/usr/bin/env node
+'use strict';
+
+/**
+ * Scaffolds a new changeset fragment (#2975).
+ *
+ *   npm run changeset -- --type Fixed --pr 1234 --body "fix the thing"
+ *
+ * Writes `.changeset/<adjective>-<noun>-<noun>.md` with frontmatter
+ * + body. The random three-word filename minimizes filename collision
+ * across concurrent PRs.
+ */
+
+const fs = require('node:fs');
+const path = require('node:path');
+
+// Small word lists — keep the function simple and dependency-free.
+// Together this gives ~40 * 40 * 40 = 64,000 distinct names. The lint
+// rejects any duplicate filename, so collisions are caught even when
+// the random draw repeats.
+const ADJECTIVES = [
+  'silly', 'brave', 'calm', 'eager', 'gentle', 'happy', 'jolly', 'kind',
+  'lively', 'merry', 'nimble', 'plucky', 'quick', 'sturdy', 'witty', 'zesty',
+  'bold', 'clever', 'daring', 'fierce', 'graceful', 'humble', 'lucky', 'noble',
+  'proud', 'rapid', 'sharp', 'tidy', 'vivid', 'wise', 'agile', 'curious',
+  'eager', 'gallant', 'mellow', 'patient', 'serene', 'steady', 'sturdy', 'sunny',
+];
+const NOUNS_A = [
+  'bears', 'birds', 'cats', 'dogs', 'elks', 'foxes', 'goats', 'hawks',
+  'ibex', 'jays', 'koalas', 'lynx', 'moles', 'newts', 'otters', 'pumas',
+  'quails', 'rams', 'seals', 'tigers', 'voles', 'wolves', 'yaks', 'zebras',
+  'badgers', 'cranes', 'deer', 'eagles', 'finches', 'geese', 'herons', 'jaguars',
+  'lemurs', 'mice', 'orcas', 'pandas', 'ravens', 'sloths', 'tunas', 'wasps',
+];
+const NOUNS_B = [
+  'dance', 'sing', 'leap', 'run', 'jump', 'climb', 'fly', 'swim',
+  'rest', 'wake', 'roam', 'greet', 'wander', 'gather', 'forage', 'travel',
+  'glide', 'sprint', 'tumble', 'wave', 'cheer', 'rally', 'parade', 'march',
+  'hop', 'frolic', 'caper', 'romp', 'zip', 'dart', 'snooze', 'munch',
+  'chatter', 'squeak', 'howl', 'bark', 'purr', 'roar', 'hum', 'click',
+];
+
+function pick(arr) {
+  return arr[Math.floor(Math.random() * arr.length)];
+}
+
+function generateFragmentName() {
+  return `${pick(ADJECTIVES)}-${pick(NOUNS_A)}-${pick(NOUNS_B)}`;
+}
+
+// Allowed Keep-a-Changelog section types. Used by both scaffoldFragment
+// (sanitization at write time) and parse.cjs (validation at consume time).
+const ALLOWED_TYPES = new Set(['Added', 'Changed', 'Deprecated', 'Removed', 'Fixed', 'Security']);
+
+function scaffoldFragment({ repo, type, pr, body }) {
+  // Sanitize: reject any type value not on the allowlist BEFORE embedding it
+  // in frontmatter. A newline in `type` would corrupt the fragment; an
+  // unrecognized value would be rejected later by parse.cjs but with a
+  // confusing diagnostic. Catch both at the write boundary.
+  if (!ALLOWED_TYPES.has(type)) {
+    throw new Error(
+      `scaffoldFragment: type=${JSON.stringify(type)} is not one of [${[...ALLOWED_TYPES].join(', ')}]`,
+    );
+  }
+  const dir = path.join(repo, '.changeset');
+  fs.mkdirSync(dir, { recursive: true });
+  const content = `---\ntype: ${type}\npr: ${pr}\n---\n${body}\n`;
+  // Atomic create: writeFileSync with `flag: 'wx'` fails (EEXIST) when the
+  // file already exists, so concurrent invocations can't race past
+  // `existsSync` and overwrite each other. Re-roll the random name on
+  // collision; fail loudly after exhausting the retry budget.
+  for (let i = 0; i < 16; i++) {
+    const name = generateFragmentName();
+    const target = path.join(dir, `${name}.md`);
+    try {
+      fs.writeFileSync(target, content, { flag: 'wx' });
+      return target;
+    } catch (e) {
+      if (e.code !== 'EEXIST') throw e;
+      // collision — try another random draw
+    }
+  }
+  throw new Error(
+    'scaffoldFragment: 16 random filename draws all collided; ' +
+    'expand the word lists or investigate corrupted .changeset/ state',
+  );
+}
+
+function parseArgs(argv) {
+  const opts = { type: null, pr: null, body: null, repo: process.cwd() };
+  // Validate flag values: argv[++i] could be undefined (flag with no value)
+  // or another flag (silently misparsed). Match the cli.cjs convention: return
+  // { ok: true, opts } on success, { ok: false, error } on malformed input.
+  const requireValue = (flag, i) => {
+    const v = argv[i + 1];
+    if (v === undefined || v.startsWith('--')) {
+      return { ok: false, error: `missing value for ${flag}` };
+    }
+    return { ok: true, value: v };
+  };
+
+  for (let i = 0; i < argv.length; i++) {
+    const a = argv[i];
+    if (a === '--type' || a === '--pr' || a === '--body' || a === '--repo') {
+      const r = requireValue(a, i);
+      if (!r.ok) return { ok: false, error: r.error };
+      if (a === '--type') opts.type = r.value;
+      else if (a === '--pr') opts.pr = Number(r.value);
+      else if (a === '--body') opts.body = r.value;
+      else if (a === '--repo') opts.repo = r.value;
+      i++;
+      continue;
+    }
+    return { ok: false, error: `unknown argument: ${a}` };
+  }
+  return { ok: true, opts };
+}
+
+function main() {
+  const parsed = parseArgs(process.argv.slice(2));
+  if (!parsed.ok) {
+    process.stderr.write(`${parsed.error}\n`);
+    process.stderr.write('usage: changeset/new.cjs --type <Fixed|Added|...> --pr NNNN --body "..."\n');
+    process.exit(2);
+  }
+  const { opts } = parsed;
+  if (!opts.type || !opts.pr || !opts.body) {
+    process.stderr.write('usage: changeset/new.cjs --type <Fixed|Added|...> --pr NNNN --body "..."\n');
+    process.exit(2);
+  }
+  const file = scaffoldFragment(opts);
+  process.stdout.write(`${path.relative(process.cwd(), file)}\n`);
+}
+
+if (require.main === module) main();
+
+module.exports = { generateFragmentName, scaffoldFragment, parseArgs, ALLOWED_TYPES };

scripts/changeset/parse.cjs

@@ -0,0 +1,60 @@
+'use strict';
+
+/**
+ * Parses a changeset fragment file (text → typed record).
+ *
+ *   ---
+ *   type: Fixed
+ *   pr: 2975
+ *   ---
+ *   <markdown body>
+ *
+ * Returns { ok: true, fragment: { type, pr, body } } on success,
+ * { ok: false, reason: FRAGMENT_ERROR.X, detail } on failure.
+ *
+ * The reason field is a frozen enum so tests assert on stable codes,
+ * not free-text error messages (CONTRIBUTING.md: "Prohibited: Raw
+ * Text Matching on Test Outputs").
+ */
+const FRAGMENT_ERROR = Object.freeze({
+  MISSING_FRONTMATTER: 'missing_frontmatter',
+  MISSING_TYPE: 'missing_type',
+  INVALID_TYPE: 'invalid_type',
+  MISSING_PR: 'missing_pr',
+  INVALID_PR: 'invalid_pr',
+  EMPTY_BODY: 'empty_body',
+});
+
+const ALLOWED_TYPES = new Set(['Added', 'Changed', 'Deprecated', 'Removed', 'Fixed', 'Security']);
+
+function parseFragment(src) {
+  const fmMatch = src.match(/^---\r?\n([\s\S]*?)\r?\n---\r?\n([\s\S]*)$/);
+  if (!fmMatch) return { ok: false, reason: FRAGMENT_ERROR.MISSING_FRONTMATTER };
+  const [, fmBlock, body] = fmMatch;
+
+  const fields = {};
+  for (const line of fmBlock.split(/\r?\n/)) {
+    const m = line.match(/^([a-zA-Z0-9_-]+):\s*(.*)$/);
+    if (m) fields[m[1]] = m[2].trim();
+  }
+
+  if (!fields.type) return { ok: false, reason: FRAGMENT_ERROR.MISSING_TYPE };
+  if (!ALLOWED_TYPES.has(fields.type)) {
+    return { ok: false, reason: FRAGMENT_ERROR.INVALID_TYPE, detail: fields.type };
+  }
+  if (!fields.pr) return { ok: false, reason: FRAGMENT_ERROR.MISSING_PR };
+  const pr = Number(fields.pr);
+  if (!Number.isInteger(pr) || pr <= 0) {
+    return { ok: false, reason: FRAGMENT_ERROR.INVALID_PR, detail: fields.pr };
+  }
+  // Use trim() only for the emptiness check; preserve the body verbatim
+  // (including significant leading/trailing whitespace, code blocks, etc.)
+  // so render → serialize round-trips exactly. Strip only a single trailing
+  // newline added by editors so byte-equality holds for typical fragments.
+  if (!body.trim()) return { ok: false, reason: FRAGMENT_ERROR.EMPTY_BODY };
+  const verbatimBody = body.endsWith('\n') ? body.slice(0, -1) : body;
+
+  return { ok: true, fragment: { type: fields.type, pr, body: verbatimBody } };
+}
+
+module.exports = { parseFragment, FRAGMENT_ERROR, ALLOWED_TYPES };

scripts/changeset/render.cjs

@@ -0,0 +1,34 @@
+'use strict';
+
+/**
+ * Pure renderer for the changeset-fragment workflow (#2975).
+ *
+ * Returns a typed Changelog IR — no file I/O. The IR is the contract that
+ * tests assert on; the markdown serializer is a separate concern.
+ *
+ *   IR shape: {
+ *     releaseHeader: { version: string, date: string },
+ *     sections: [{ type: string, bullets: [{ pr: number, body: string }] }],
+ *     priorChangelog: string | null,
+ *   }
+ */
+// Keep a Changelog (https://keepachangelog.com) standard section order.
+const SECTION_ORDER = ['Added', 'Changed', 'Deprecated', 'Removed', 'Fixed', 'Security'];
+
+function renderChangelog({ fragments, version, date, priorChangelog }) {
+  const byType = new Map();
+  for (const f of fragments) {
+    if (!byType.has(f.type)) byType.set(f.type, []);
+    byType.get(f.type).push({ pr: f.pr, body: f.body });
+  }
+  const sections = SECTION_ORDER
+    .filter((type) => byType.has(type))
+    .map((type) => ({ type, bullets: byType.get(type) }));
+  return {
+    releaseHeader: { version, date },
+    sections,
+    priorChangelog: priorChangelog || null,
+  };
+}
+
+module.exports = { renderChangelog };

scripts/changeset/serialize.cjs

@@ -0,0 +1,74 @@
+'use strict';
+
+/**
+ * Markdown serializer + parser for the changelog IR. The two are inverses
+ * over the well-formed subset; tests assert via round-trip (parse(serialize(ir)))
+ * rather than by inspecting serialized text — see CONTRIBUTING.md
+ * "Prohibited: Raw Text Matching on Test Outputs".
+ *
+ * Serialized form (Keep a Changelog):
+ *
+ *   ## [1.42.0] - 2026-05-01
+ *
+ *   ### Fixed
+ *
+ *   - body of the bullet (#NNNN)
+ *
+ *   <priorChangelog appended verbatim>
+ */
+
+function serializeChangelog(ir) {
+  const lines = [];
+  const { version, date } = ir.releaseHeader;
+  lines.push(`## [${version}] - ${date}`);
+  lines.push('');
+  for (const section of ir.sections) {
+    lines.push(`### ${section.type}`);
+    lines.push('');
+    for (const b of section.bullets) {
+      lines.push(`- ${b.body} (#${b.pr})`);
+    }
+    lines.push('');
+  }
+  let out = lines.join('\n');
+  if (ir.priorChangelog) {
+    out += '\n' + ir.priorChangelog;
+  }
+  return out;
+}
+
+/**
+ * Inverse parser: extracts the structured releases from a CHANGELOG.md
+ * text. Returns { releases: [{ version, date, sections: [{ type, bullets:
+ * [{ pr, body }] }] }] }. Tolerates the actual repo's CHANGELOG dialect.
+ */
+function parseChangelog(text) {
+  const releases = [];
+  const lines = text.split(/\r?\n/);
+  let cur = null;
+  let curSection = null;
+  for (const line of lines) {
+    const releaseMatch = line.match(/^##\s+\[([^\]]+)\](?:\s*-\s*(\S+))?/);
+    if (releaseMatch) {
+      cur = { version: releaseMatch[1], date: releaseMatch[2] || null, sections: [] };
+      curSection = null;
+      releases.push(cur);
+      continue;
+    }
+    if (!cur) continue;
+    const sectionMatch = line.match(/^###\s+(.+?)\s*$/);
+    if (sectionMatch) {
+      curSection = { type: sectionMatch[1], bullets: [] };
+      cur.sections.push(curSection);
+      continue;
+    }
+    if (!curSection) continue;
+    const bulletMatch = line.match(/^-\s+(.*?)\s*\(#(\d+)\)\s*$/);
+    if (bulletMatch) {
+      curSection.bullets.push({ body: bulletMatch[1], pr: Number(bulletMatch[2]) });
+    }
+  }
+  return { releases };
+}
+
+module.exports = { serializeChangelog, parseChangelog };