test(scripts): pin the published stylesheet banner per subject (#7044)

[yinlianghui]

Fixes#7044

What

Three packages ship a supplement stylesheet whose emitted bytes open with a
banner comment that is part of the published artifact (@object-ui/fields,
@object-ui/plugin-grid, @object-ui/plugin-kanban). Nothing in the suite
read that banner, so a change to the shared defaultHeader, or an accidental
drop of fields' own per-package header override, would ship silently — the
second case being a diff in a published file that #6405's acceptance gate
(byte-identity of the emitted sheet) was explicitly meant to prevent.

This PR:

• Exports defaultHeader from scripts/build-plugin-stylesheet.mjs (was
  module-private; one-line change, function body untouched, plus a doc-comment
  note on why it's exported now).
• Adds a per-subject assertion in
  scripts/__tests__/plugin-published-stylesheet.test.ts that the emitted
  sheet opens with the header the subject is actually configured to use —
  its own declared buildOptions.header for fields, else the shared
  defaultHeader(PACKAGE_NAME) this module now exports for plugin-grid /
  plugin-kanban — never a second, hand-spelled copy of either wording (the
  route ruled out on the issue as the "assertion that inspects nothing"
  shape).

Why two checks, not one

The obvious single assertion — recompute "expected" by calling the very
function (defaultHeader, or buildOptions.header) that also produced
"actual" — is vacuous: since both sides call the same pure function on the
same input, no mutation to that function's own wording can ever make it
disagree with itself. Ablating confirmed this (see Testing below), so the
assertion adds two independent checks that don't have that shape:

1. Assembly check (all 3 subjects): the emitted sheet must start with
   expectedHeader, where expectedHeader is read from the subject's own
   build script export. Catches the header not being prepended at all, or
   some unrelated value being used.
2. Name-presence check (all 3 subjects): the banner region contains the
   subject's own PACKAGE_NAME — a plain string constant each build script
   declares independently, never derived from defaultHeader. This is what
   gives check 1 teeth against a broken interpolation inside defaultHeader
   itself, for the two packages that inherit it.
3. fields-only control: fields' emitted sheet must NOT start with the
   shared default banner. This is the check that actually catches "an
   accidental drop of fields' header override" — the scenario named in the
   issue — because dropping it moves fields' "expected" (check 1's fallback)
   and "actual" to the shared default in lockstep, so check 1 alone can't see
   it.

Testing

pnpm exec vitest run scripts/__tests__/plugin-published-stylesheet.test.ts --maxWorkers=2
# Test Files 1 passed (1) / Tests 24 passed (24)
pnpm type-check:scripts # clean
pnpm check:control-bytes # OK, 6004 tracked text files scanned
pnpm check:entry-guard # 52 export bindings, 52 inert on import (0 known-unsafe)
pnpm check:esm-specifiers # unaffected (no ESM specifier surface touched)
node scripts/check-changeset-presence.mjs # no changeset owed (scripts/-only diff)

Ablation 1 — mutated defaultHeader() in
scripts/build-plugin-stylesheet.mjs to hardcode a wrong package name
(breaking its package-name interpolation), confirmed the mutation landed on
disk (git diff, byte count), reran the suite:

FAIL plugin-grid > opens the emitted sheet with the banner it declares
FAIL plugin-kanban > opens the emitted sheet with the banner it declares
PASS fields > opens the emitted sheet with the banner it declares
PASS fields > does not fall back to the shared default banner
Tests 2 failed | 22 passed (24)

Restored the mutated file from HEAD (git checkout HEAD on that one path);
git hash-object on the restored file matched the pre-mutation blob exactly.

Ablation 2 — mutated packages/fields/scripts/build-css.mjs to drop its
header: HEADER, line from buildOptions (simulating the exact "accidental
drop" defect named in the issue), confirmed on disk, reran the suite:

PASS plugin-grid > opens the emitted sheet with the banner it declares
PASS plugin-kanban > opens the emitted sheet with the banner it declares
PASS fields > opens the emitted sheet with the banner it declares (self-consistent — expected)
FAIL fields > does not fall back to the shared default banner
Tests 1 failed | 23 passed (24)

Restored the mutated file from HEAD the same way; git hash-object matched
the pre-mutation blob exactly.

Byte-identity of the published artifact (objectui#6405's acceptance
gate): built all three subjects' emitted sheets against this branch's
scripts/build-plugin-stylesheet.mjs and, separately, against the pre-PR
version at 64d624ded (temporarily checked out, restored by blob hash
afterward). SHA-256 of all three emitted sheets is identical before/after —
the export keyword and doc-comment addition change zero published bytes.

Scope

Touches only scripts/build-plugin-stylesheet.mjs (the one-line export +
a doc-comment) and scripts/__tests__/plugin-published-stylesheet.test.ts.
The three build-css.mjs consumers are untouched in the committed diff (they
were only mutated transiently, on disk, during the two ablations above, and
restored by verified blob hash — never part of any commit).

No changeset: scripts-only diff, nothing published moves.

🤖 Generated with Claude Code

https://claude.ai/code/session_01BGMDbrVa8JjZcCQ7DWYH1b

[yinlianghui]

Fixes#7044

What

Three packages ship a supplement stylesheet whose emitted bytes open with a
banner comment that is part of the published artifact (@object-ui/fields,
@object-ui/plugin-grid, @object-ui/plugin-kanban). Nothing in the suite
read that banner, so a change to the shared defaultHeader, or an accidental
drop of fields' own per-package header override, would ship silently — the
second case being a diff in a published file that #6405's acceptance gate
(byte-identity of the emitted sheet) was explicitly meant to prevent.

This PR:

• Exports defaultHeader from scripts/build-plugin-stylesheet.mjs (was
  module-private; one-line change, function body untouched, plus a doc-comment
  note on why it's exported now).
• Adds a per-subject assertion in
  scripts/__tests__/plugin-published-stylesheet.test.ts that the emitted
  sheet opens with the header the subject is actually configured to use —
  its own declared buildOptions.header for fields, else the shared
  defaultHeader(PACKAGE_NAME) this module now exports for plugin-grid /
  plugin-kanban — never a second, hand-spelled copy of either wording (the
  route ruled out on the issue as the "assertion that inspects nothing"
  shape).

Why two checks, not one

The obvious single assertion — recompute "expected" by calling the very
function (defaultHeader, or buildOptions.header) that also produced
"actual" — is vacuous: since both sides call the same pure function on the
same input, no mutation to that function's own wording can ever make it
disagree with itself. Ablating confirmed this (see Testing below), so the
assertion adds two independent checks that don't have that shape:

1. Assembly check (all 3 subjects): the emitted sheet must start with
   expectedHeader, where expectedHeader is read from the subject's own
   build script export. Catches the header not being prepended at all, or
   some unrelated value being used.
2. Name-presence check (all 3 subjects): the banner region contains the
   subject's own PACKAGE_NAME — a plain string constant each build script
   declares independently, never derived from defaultHeader. This is what
   gives check 1 teeth against a broken interpolation inside defaultHeader
   itself, for the two packages that inherit it.
3. fields-only control: fields' emitted sheet must NOT start with the
   shared default banner. This is the check that actually catches "an
   accidental drop of fields' header override" — the scenario named in the
   issue — because dropping it moves fields' "expected" (check 1's fallback)
   and "actual" to the shared default in lockstep, so check 1 alone can't see
   it.

Testing

pnpm exec vitest run scripts/__tests__/plugin-published-stylesheet.test.ts --maxWorkers=2
# Test Files 1 passed (1) / Tests 24 passed (24)
pnpm type-check:scripts # clean
pnpm check:control-bytes # OK, 6004 tracked text files scanned
pnpm check:entry-guard # 52 export bindings, 52 inert on import (0 known-unsafe)
pnpm check:esm-specifiers # unaffected (no ESM specifier surface touched)
node scripts/check-changeset-presence.mjs # no changeset owed (scripts/-only diff)

Ablation 1 — mutated defaultHeader() in
scripts/build-plugin-stylesheet.mjs to hardcode a wrong package name
(breaking its package-name interpolation), confirmed the mutation landed on
disk (git diff, byte count), reran the suite:

FAIL plugin-grid > opens the emitted sheet with the banner it declares
FAIL plugin-kanban > opens the emitted sheet with the banner it declares
PASS fields > opens the emitted sheet with the banner it declares
PASS fields > does not fall back to the shared default banner
Tests 2 failed | 22 passed (24)

Restored the mutated file from HEAD (git checkout HEAD on that one path);
git hash-object on the restored file matched the pre-mutation blob exactly.

Ablation 2 — mutated packages/fields/scripts/build-css.mjs to drop its
header: HEADER, line from buildOptions (simulating the exact "accidental
drop" defect named in the issue), confirmed on disk, reran the suite:

PASS plugin-grid > opens the emitted sheet with the banner it declares
PASS plugin-kanban > opens the emitted sheet with the banner it declares
PASS fields > opens the emitted sheet with the banner it declares (self-consistent — expected)
FAIL fields > does not fall back to the shared default banner
Tests 1 failed | 23 passed (24)

Restored the mutated file from HEAD the same way; git hash-object matched
the pre-mutation blob exactly.

Byte-identity of the published artifact (objectui#6405's acceptance
gate): built all three subjects' emitted sheets against this branch's
scripts/build-plugin-stylesheet.mjs and, separately, against the pre-PR
version at 64d624ded (temporarily checked out, restored by blob hash
afterward). SHA-256 of all three emitted sheets is identical before/after —
the export keyword and doc-comment addition change zero published bytes.

Scope

Touches only scripts/build-plugin-stylesheet.mjs (the one-line export +
a doc-comment) and scripts/__tests__/plugin-published-stylesheet.test.ts.
The three build-css.mjs consumers are untouched in the committed diff (they
were only mutated transiently, on disk, during the two ablations above, and
restored by verified blob hash — never part of any commit).

No changeset: scripts-only diff, nothing published moves.

🤖 Generated with Claude Code

https://claude.ai/code/session_01BGMDbrVa8JjZcCQ7DWYH1b

[yinlianghui]

Fixes#7044

What

Three packages ship a supplement stylesheet whose emitted bytes open with a
banner comment that is part of the published artifact (@object-ui/fields,
@object-ui/plugin-grid, @object-ui/plugin-kanban). Nothing in the suite
read that banner, so a change to the shared defaultHeader, or an accidental
drop of fields' own per-package header override, would ship silently — the
second case being a diff in a published file that #6405's acceptance gate
(byte-identity of the emitted sheet) was explicitly meant to prevent.

This PR:

• Exports defaultHeader from scripts/build-plugin-stylesheet.mjs (was
  module-private; one-line change, function body untouched, plus a doc-comment
  note on why it's exported now).
• Adds a per-subject assertion in
  scripts/__tests__/plugin-published-stylesheet.test.ts that the emitted
  sheet opens with the header the subject is actually configured to use —
  its own declared buildOptions.header for fields, else the shared
  defaultHeader(PACKAGE_NAME) this module now exports for plugin-grid /
  plugin-kanban — never a second, hand-spelled copy of either wording (the
  route ruled out on the issue as the "assertion that inspects nothing"
  shape).

Why two checks, not one

The obvious single assertion — recompute "expected" by calling the very
function (defaultHeader, or buildOptions.header) that also produced
"actual" — is vacuous: since both sides call the same pure function on the
same input, no mutation to that function's own wording can ever make it
disagree with itself. Ablating confirmed this (see Testing below), so the
assertion adds two independent checks that don't have that shape:

1. Assembly check (all 3 subjects): the emitted sheet must start with
   expectedHeader, where expectedHeader is read from the subject's own
   build script export. Catches the header not being prepended at all, or
   some unrelated value being used.
2. Name-presence check (all 3 subjects): the banner region contains the
   subject's own PACKAGE_NAME — a plain string constant each build script
   declares independently, never derived from defaultHeader. This is what
   gives check 1 teeth against a broken interpolation inside defaultHeader
   itself, for the two packages that inherit it.
3. fields-only control: fields' emitted sheet must NOT start with the
   shared default banner. This is the check that actually catches "an
   accidental drop of fields' header override" — the scenario named in the
   issue — because dropping it moves fields' "expected" (check 1's fallback)
   and "actual" to the shared default in lockstep, so check 1 alone can't see
   it.

Testing

pnpm exec vitest run scripts/__tests__/plugin-published-stylesheet.test.ts --maxWorkers=2
# Test Files 1 passed (1) / Tests 24 passed (24)
pnpm type-check:scripts # clean
pnpm check:control-bytes # OK, 6004 tracked text files scanned
pnpm check:entry-guard # 52 export bindings, 52 inert on import (0 known-unsafe)
pnpm check:esm-specifiers # unaffected (no ESM specifier surface touched)
node scripts/check-changeset-presence.mjs # no changeset owed (scripts/-only diff)

Ablation 1 — mutated defaultHeader() in
scripts/build-plugin-stylesheet.mjs to hardcode a wrong package name
(breaking its package-name interpolation), confirmed the mutation landed on
disk (git diff, byte count), reran the suite:

FAIL plugin-grid > opens the emitted sheet with the banner it declares
FAIL plugin-kanban > opens the emitted sheet with the banner it declares
PASS fields > opens the emitted sheet with the banner it declares
PASS fields > does not fall back to the shared default banner
Tests 2 failed | 22 passed (24)

Restored the mutated file from HEAD (git checkout HEAD on that one path);
git hash-object on the restored file matched the pre-mutation blob exactly.

Ablation 2 — mutated packages/fields/scripts/build-css.mjs to drop its
header: HEADER, line from buildOptions (simulating the exact "accidental
drop" defect named in the issue), confirmed on disk, reran the suite:

PASS plugin-grid > opens the emitted sheet with the banner it declares
PASS plugin-kanban > opens the emitted sheet with the banner it declares
PASS fields > opens the emitted sheet with the banner it declares (self-consistent — expected)
FAIL fields > does not fall back to the shared default banner
Tests 1 failed | 23 passed (24)

Restored the mutated file from HEAD the same way; git hash-object matched
the pre-mutation blob exactly.

Byte-identity of the published artifact (objectui#6405's acceptance
gate): built all three subjects' emitted sheets against this branch's
scripts/build-plugin-stylesheet.mjs and, separately, against the pre-PR
version at 64d624ded (temporarily checked out, restored by blob hash
afterward). SHA-256 of all three emitted sheets is identical before/after —
the export keyword and doc-comment addition change zero published bytes.

Scope

Touches only scripts/build-plugin-stylesheet.mjs (the one-line export +
a doc-comment) and scripts/__tests__/plugin-published-stylesheet.test.ts.
The three build-css.mjs consumers are untouched in the committed diff (they
were only mutated transiently, on disk, during the two ablations above, and
restored by verified blob hash — never part of any commit).

No changeset: scripts-only diff, nothing published moves.

🤖 Generated with Claude Code

https://claude.ai/code/session_01BGMDbrVa8JjZcCQ7DWYH1b

[yinlianghui]

Fixes#7044

What

Three packages ship a supplement stylesheet whose emitted bytes open with a
banner comment that is part of the published artifact (@object-ui/fields,
@object-ui/plugin-grid, @object-ui/plugin-kanban). Nothing in the suite
read that banner, so a change to the shared defaultHeader, or an accidental
drop of fields' own per-package header override, would ship silently — the
second case being a diff in a published file that #6405's acceptance gate
(byte-identity of the emitted sheet) was explicitly meant to prevent.

This PR:

• Exports defaultHeader from scripts/build-plugin-stylesheet.mjs (was
  module-private; one-line change, function body untouched, plus a doc-comment
  note on why it's exported now).
• Adds a per-subject assertion in
  scripts/__tests__/plugin-published-stylesheet.test.ts that the emitted
  sheet opens with the header the subject is actually configured to use —
  its own declared buildOptions.header for fields, else the shared
  defaultHeader(PACKAGE_NAME) this module now exports for plugin-grid /
  plugin-kanban — never a second, hand-spelled copy of either wording (the
  route ruled out on the issue as the "assertion that inspects nothing"
  shape).

Why two checks, not one

The obvious single assertion — recompute "expected" by calling the very
function (defaultHeader, or buildOptions.header) that also produced
"actual" — is vacuous: since both sides call the same pure function on the
same input, no mutation to that function's own wording can ever make it
disagree with itself. Ablating confirmed this (see Testing below), so the
assertion adds two independent checks that don't have that shape:

1. Assembly check (all 3 subjects): the emitted sheet must start with
   expectedHeader, where expectedHeader is read from the subject's own
   build script export. Catches the header not being prepended at all, or
   some unrelated value being used.
2. Name-presence check (all 3 subjects): the banner region contains the
   subject's own PACKAGE_NAME — a plain string constant each build script
   declares independently, never derived from defaultHeader. This is what
   gives check 1 teeth against a broken interpolation inside defaultHeader
   itself, for the two packages that inherit it.
3. fields-only control: fields' emitted sheet must NOT start with the
   shared default banner. This is the check that actually catches "an
   accidental drop of fields' header override" — the scenario named in the
   issue — because dropping it moves fields' "expected" (check 1's fallback)
   and "actual" to the shared default in lockstep, so check 1 alone can't see
   it.

Testing

pnpm exec vitest run scripts/__tests__/plugin-published-stylesheet.test.ts --maxWorkers=2
# Test Files 1 passed (1) / Tests 24 passed (24)
pnpm type-check:scripts # clean
pnpm check:control-bytes # OK, 6004 tracked text files scanned
pnpm check:entry-guard # 52 export bindings, 52 inert on import (0 known-unsafe)
pnpm check:esm-specifiers # unaffected (no ESM specifier surface touched)
node scripts/check-changeset-presence.mjs # no changeset owed (scripts/-only diff)

Ablation 1 — mutated defaultHeader() in
scripts/build-plugin-stylesheet.mjs to hardcode a wrong package name
(breaking its package-name interpolation), confirmed the mutation landed on
disk (git diff, byte count), reran the suite:

FAIL plugin-grid > opens the emitted sheet with the banner it declares
FAIL plugin-kanban > opens the emitted sheet with the banner it declares
PASS fields > opens the emitted sheet with the banner it declares
PASS fields > does not fall back to the shared default banner
Tests 2 failed | 22 passed (24)

Restored the mutated file from HEAD (git checkout HEAD on that one path);
git hash-object on the restored file matched the pre-mutation blob exactly.

Ablation 2 — mutated packages/fields/scripts/build-css.mjs to drop its
header: HEADER, line from buildOptions (simulating the exact "accidental
drop" defect named in the issue), confirmed on disk, reran the suite:

PASS plugin-grid > opens the emitted sheet with the banner it declares
PASS plugin-kanban > opens the emitted sheet with the banner it declares
PASS fields > opens the emitted sheet with the banner it declares (self-consistent — expected)
FAIL fields > does not fall back to the shared default banner
Tests 1 failed | 23 passed (24)

Restored the mutated file from HEAD the same way; git hash-object matched
the pre-mutation blob exactly.

Byte-identity of the published artifact (objectui#6405's acceptance
gate): built all three subjects' emitted sheets against this branch's
scripts/build-plugin-stylesheet.mjs and, separately, against the pre-PR
version at 64d624ded (temporarily checked out, restored by blob hash
afterward). SHA-256 of all three emitted sheets is identical before/after —
the export keyword and doc-comment addition change zero published bytes.

Scope

Touches only scripts/build-plugin-stylesheet.mjs (the one-line export +
a doc-comment) and scripts/__tests__/plugin-published-stylesheet.test.ts.
The three build-css.mjs consumers are untouched in the committed diff (they
were only mutated transiently, on disk, during the two ablations above, and
restored by verified blob hash — never part of any commit).

No changeset: scripts-only diff, nothing published moves.

🤖 Generated with Claude Code

https://claude.ai/code/session_01BGMDbrVa8JjZcCQ7DWYH1b

[yinlianghui]

Fixes#7044

What

Three packages ship a supplement stylesheet whose emitted bytes open with a
banner comment that is part of the published artifact (@object-ui/fields,
@object-ui/plugin-grid, @object-ui/plugin-kanban). Nothing in the suite
read that banner, so a change to the shared defaultHeader, or an accidental
drop of fields' own per-package header override, would ship silently — the
second case being a diff in a published file that #6405's acceptance gate
(byte-identity of the emitted sheet) was explicitly meant to prevent.

This PR:

• Exports defaultHeader from scripts/build-plugin-stylesheet.mjs (was
  module-private; one-line change, function body untouched, plus a doc-comment
  note on why it's exported now).
• Adds a per-subject assertion in
  scripts/__tests__/plugin-published-stylesheet.test.ts that the emitted
  sheet opens with the header the subject is actually configured to use —
  its own declared buildOptions.header for fields, else the shared
  defaultHeader(PACKAGE_NAME) this module now exports for plugin-grid /
  plugin-kanban — never a second, hand-spelled copy of either wording (the
  route ruled out on the issue as the "assertion that inspects nothing"
  shape).

Why two checks, not one

The obvious single assertion — recompute "expected" by calling the very
function (defaultHeader, or buildOptions.header) that also produced
"actual" — is vacuous: since both sides call the same pure function on the
same input, no mutation to that function's own wording can ever make it
disagree with itself. Ablating confirmed this (see Testing below), so the
assertion adds two independent checks that don't have that shape:

1. Assembly check (all 3 subjects): the emitted sheet must start with
   expectedHeader, where expectedHeader is read from the subject's own
   build script export. Catches the header not being prepended at all, or
   some unrelated value being used.
2. Name-presence check (all 3 subjects): the banner region contains the
   subject's own PACKAGE_NAME — a plain string constant each build script
   declares independently, never derived from defaultHeader. This is what
   gives check 1 teeth against a broken interpolation inside defaultHeader
   itself, for the two packages that inherit it.
3. fields-only control: fields' emitted sheet must NOT start with the
   shared default banner. This is the check that actually catches "an
   accidental drop of fields' header override" — the scenario named in the
   issue — because dropping it moves fields' "expected" (check 1's fallback)
   and "actual" to the shared default in lockstep, so check 1 alone can't see
   it.

Testing

pnpm exec vitest run scripts/__tests__/plugin-published-stylesheet.test.ts --maxWorkers=2
# Test Files 1 passed (1) / Tests 24 passed (24)
pnpm type-check:scripts # clean
pnpm check:control-bytes # OK, 6004 tracked text files scanned
pnpm check:entry-guard # 52 export bindings, 52 inert on import (0 known-unsafe)
pnpm check:esm-specifiers # unaffected (no ESM specifier surface touched)
node scripts/check-changeset-presence.mjs # no changeset owed (scripts/-only diff)

Ablation 1 — mutated defaultHeader() in
scripts/build-plugin-stylesheet.mjs to hardcode a wrong package name
(breaking its package-name interpolation), confirmed the mutation landed on
disk (git diff, byte count), reran the suite:

FAIL plugin-grid > opens the emitted sheet with the banner it declares
FAIL plugin-kanban > opens the emitted sheet with the banner it declares
PASS fields > opens the emitted sheet with the banner it declares
PASS fields > does not fall back to the shared default banner
Tests 2 failed | 22 passed (24)

Restored the mutated file from HEAD (git checkout HEAD on that one path);
git hash-object on the restored file matched the pre-mutation blob exactly.

Ablation 2 — mutated packages/fields/scripts/build-css.mjs to drop its
header: HEADER, line from buildOptions (simulating the exact "accidental
drop" defect named in the issue), confirmed on disk, reran the suite:

PASS plugin-grid > opens the emitted sheet with the banner it declares
PASS plugin-kanban > opens the emitted sheet with the banner it declares
PASS fields > opens the emitted sheet with the banner it declares (self-consistent — expected)
FAIL fields > does not fall back to the shared default banner
Tests 1 failed | 23 passed (24)

Restored the mutated file from HEAD the same way; git hash-object matched
the pre-mutation blob exactly.

Byte-identity of the published artifact (objectui#6405's acceptance
gate): built all three subjects' emitted sheets against this branch's
scripts/build-plugin-stylesheet.mjs and, separately, against the pre-PR
version at 64d624ded (temporarily checked out, restored by blob hash
afterward). SHA-256 of all three emitted sheets is identical before/after —
the export keyword and doc-comment addition change zero published bytes.

Scope

Touches only scripts/build-plugin-stylesheet.mjs (the one-line export +
a doc-comment) and scripts/__tests__/plugin-published-stylesheet.test.ts.
The three build-css.mjs consumers are untouched in the committed diff (they
were only mutated transiently, on disk, during the two ablations above, and
restored by verified blob hash — never part of any commit).

No changeset: scripts-only diff, nothing published moves.

🤖 Generated with Claude Code

https://claude.ai/code/session_01BGMDbrVa8JjZcCQ7DWYH1b

[yinlianghui]

Fixes#7044

What

Three packages ship a supplement stylesheet whose emitted bytes open with a
banner comment that is part of the published artifact (@object-ui/fields,
@object-ui/plugin-grid, @object-ui/plugin-kanban). Nothing in the suite
read that banner, so a change to the shared defaultHeader, or an accidental
drop of fields' own per-package header override, would ship silently — the
second case being a diff in a published file that #6405's acceptance gate
(byte-identity of the emitted sheet) was explicitly meant to prevent.

This PR:

• Exports defaultHeader from scripts/build-plugin-stylesheet.mjs (was
  module-private; one-line change, function body untouched, plus a doc-comment
  note on why it's exported now).
• Adds a per-subject assertion in
  scripts/__tests__/plugin-published-stylesheet.test.ts that the emitted
  sheet opens with the header the subject is actually configured to use —
  its own declared buildOptions.header for fields, else the shared
  defaultHeader(PACKAGE_NAME) this module now exports for plugin-grid /
  plugin-kanban — never a second, hand-spelled copy of either wording (the
  route ruled out on the issue as the "assertion that inspects nothing"
  shape).

Why two checks, not one

The obvious single assertion — recompute "expected" by calling the very
function (defaultHeader, or buildOptions.header) that also produced
"actual" — is vacuous: since both sides call the same pure function on the
same input, no mutation to that function's own wording can ever make it
disagree with itself. Ablating confirmed this (see Testing below), so the
assertion adds two independent checks that don't have that shape:

1. Assembly check (all 3 subjects): the emitted sheet must start with
   expectedHeader, where expectedHeader is read from the subject's own
   build script export. Catches the header not being prepended at all, or
   some unrelated value being used.
2. Name-presence check (all 3 subjects): the banner region contains the
   subject's own PACKAGE_NAME — a plain string constant each build script
   declares independently, never derived from defaultHeader. This is what
   gives check 1 teeth against a broken interpolation inside defaultHeader
   itself, for the two packages that inherit it.
3. fields-only control: fields' emitted sheet must NOT start with the
   shared default banner. This is the check that actually catches "an
   accidental drop of fields' header override" — the scenario named in the
   issue — because dropping it moves fields' "expected" (check 1's fallback)
   and "actual" to the shared default in lockstep, so check 1 alone can't see
   it.

Testing

pnpm exec vitest run scripts/__tests__/plugin-published-stylesheet.test.ts --maxWorkers=2
# Test Files 1 passed (1) / Tests 24 passed (24)
pnpm type-check:scripts # clean
pnpm check:control-bytes # OK, 6004 tracked text files scanned
pnpm check:entry-guard # 52 export bindings, 52 inert on import (0 known-unsafe)
pnpm check:esm-specifiers # unaffected (no ESM specifier surface touched)
node scripts/check-changeset-presence.mjs # no changeset owed (scripts/-only diff)

Ablation 1 — mutated defaultHeader() in
scripts/build-plugin-stylesheet.mjs to hardcode a wrong package name
(breaking its package-name interpolation), confirmed the mutation landed on
disk (git diff, byte count), reran the suite:

FAIL plugin-grid > opens the emitted sheet with the banner it declares
FAIL plugin-kanban > opens the emitted sheet with the banner it declares
PASS fields > opens the emitted sheet with the banner it declares
PASS fields > does not fall back to the shared default banner
Tests 2 failed | 22 passed (24)

Restored the mutated file from HEAD (git checkout HEAD on that one path);
git hash-object on the restored file matched the pre-mutation blob exactly.

Ablation 2 — mutated packages/fields/scripts/build-css.mjs to drop its
header: HEADER, line from buildOptions (simulating the exact "accidental
drop" defect named in the issue), confirmed on disk, reran the suite:

PASS plugin-grid > opens the emitted sheet with the banner it declares
PASS plugin-kanban > opens the emitted sheet with the banner it declares
PASS fields > opens the emitted sheet with the banner it declares (self-consistent — expected)
FAIL fields > does not fall back to the shared default banner
Tests 1 failed | 23 passed (24)

Restored the mutated file from HEAD the same way; git hash-object matched
the pre-mutation blob exactly.

Byte-identity of the published artifact (objectui#6405's acceptance
gate): built all three subjects' emitted sheets against this branch's
scripts/build-plugin-stylesheet.mjs and, separately, against the pre-PR
version at 64d624ded (temporarily checked out, restored by blob hash
afterward). SHA-256 of all three emitted sheets is identical before/after —
the export keyword and doc-comment addition change zero published bytes.

Scope

Touches only scripts/build-plugin-stylesheet.mjs (the one-line export +
a doc-comment) and scripts/__tests__/plugin-published-stylesheet.test.ts.
The three build-css.mjs consumers are untouched in the committed diff (they
were only mutated transiently, on disk, during the two ablations above, and
restored by verified blob hash — never part of any commit).

No changeset: scripts-only diff, nothing published moves.

🤖 Generated with Claude Code

https://claude.ai/code/session_01BGMDbrVa8JjZcCQ7DWYH1b

[yinlianghui]

Fixes#7044

What

Three packages ship a supplement stylesheet whose emitted bytes open with a
banner comment that is part of the published artifact (@object-ui/fields,
@object-ui/plugin-grid, @object-ui/plugin-kanban). Nothing in the suite
read that banner, so a change to the shared defaultHeader, or an accidental
drop of fields' own per-package header override, would ship silently — the
second case being a diff in a published file that #6405's acceptance gate
(byte-identity of the emitted sheet) was explicitly meant to prevent.

This PR:

• Exports defaultHeader from scripts/build-plugin-stylesheet.mjs (was
  module-private; one-line change, function body untouched, plus a doc-comment
  note on why it's exported now).
• Adds a per-subject assertion in
  scripts/__tests__/plugin-published-stylesheet.test.ts that the emitted
  sheet opens with the header the subject is actually configured to use —
  its own declared buildOptions.header for fields, else the shared
  defaultHeader(PACKAGE_NAME) this module now exports for plugin-grid /
  plugin-kanban — never a second, hand-spelled copy of either wording (the
  route ruled out on the issue as the "assertion that inspects nothing"
  shape).

Why two checks, not one

The obvious single assertion — recompute "expected" by calling the very
function (defaultHeader, or buildOptions.header) that also produced
"actual" — is vacuous: since both sides call the same pure function on the
same input, no mutation to that function's own wording can ever make it
disagree with itself. Ablating confirmed this (see Testing below), so the
assertion adds two independent checks that don't have that shape:

1. Assembly check (all 3 subjects): the emitted sheet must start with
   expectedHeader, where expectedHeader is read from the subject's own
   build script export. Catches the header not being prepended at all, or
   some unrelated value being used.
2. Name-presence check (all 3 subjects): the banner region contains the
   subject's own PACKAGE_NAME — a plain string constant each build script
   declares independently, never derived from defaultHeader. This is what
   gives check 1 teeth against a broken interpolation inside defaultHeader
   itself, for the two packages that inherit it.
3. fields-only control: fields' emitted sheet must NOT start with the
   shared default banner. This is the check that actually catches "an
   accidental drop of fields' header override" — the scenario named in the
   issue — because dropping it moves fields' "expected" (check 1's fallback)
   and "actual" to the shared default in lockstep, so check 1 alone can't see
   it.

Testing

pnpm exec vitest run scripts/__tests__/plugin-published-stylesheet.test.ts --maxWorkers=2
# Test Files 1 passed (1) / Tests 24 passed (24)
pnpm type-check:scripts # clean
pnpm check:control-bytes # OK, 6004 tracked text files scanned
pnpm check:entry-guard # 52 export bindings, 52 inert on import (0 known-unsafe)
pnpm check:esm-specifiers # unaffected (no ESM specifier surface touched)
node scripts/check-changeset-presence.mjs # no changeset owed (scripts/-only diff)

Ablation 1 — mutated defaultHeader() in
scripts/build-plugin-stylesheet.mjs to hardcode a wrong package name
(breaking its package-name interpolation), confirmed the mutation landed on
disk (git diff, byte count), reran the suite:

FAIL plugin-grid > opens the emitted sheet with the banner it declares
FAIL plugin-kanban > opens the emitted sheet with the banner it declares
PASS fields > opens the emitted sheet with the banner it declares
PASS fields > does not fall back to the shared default banner
Tests 2 failed | 22 passed (24)

Restored the mutated file from HEAD (git checkout HEAD on that one path);
git hash-object on the restored file matched the pre-mutation blob exactly.

Ablation 2 — mutated packages/fields/scripts/build-css.mjs to drop its
header: HEADER, line from buildOptions (simulating the exact "accidental
drop" defect named in the issue), confirmed on disk, reran the suite:

PASS plugin-grid > opens the emitted sheet with the banner it declares
PASS plugin-kanban > opens the emitted sheet with the banner it declares
PASS fields > opens the emitted sheet with the banner it declares (self-consistent — expected)
FAIL fields > does not fall back to the shared default banner
Tests 1 failed | 23 passed (24)

Restored the mutated file from HEAD the same way; git hash-object matched
the pre-mutation blob exactly.

Byte-identity of the published artifact (objectui#6405's acceptance
gate): built all three subjects' emitted sheets against this branch's
scripts/build-plugin-stylesheet.mjs and, separately, against the pre-PR
version at 64d624ded (temporarily checked out, restored by blob hash
afterward). SHA-256 of all three emitted sheets is identical before/after —
the export keyword and doc-comment addition change zero published bytes.

Scope

Touches only scripts/build-plugin-stylesheet.mjs (the one-line export +
a doc-comment) and scripts/__tests__/plugin-published-stylesheet.test.ts.
The three build-css.mjs consumers are untouched in the committed diff (they
were only mutated transiently, on disk, during the two ablations above, and
restored by verified blob hash — never part of any commit).

No changeset: scripts-only diff, nothing published moves.

🤖 Generated with Claude Code

https://claude.ai/code/session_01BGMDbrVa8JjZcCQ7DWYH1b

[yinlianghui]

Fixes#7044

What

Three packages ship a supplement stylesheet whose emitted bytes open with a
banner comment that is part of the published artifact (@object-ui/fields,
@object-ui/plugin-grid, @object-ui/plugin-kanban). Nothing in the suite
read that banner, so a change to the shared defaultHeader, or an accidental
drop of fields' own per-package header override, would ship silently — the
second case being a diff in a published file that #6405's acceptance gate
(byte-identity of the emitted sheet) was explicitly meant to prevent.

This PR:

• Exports defaultHeader from scripts/build-plugin-stylesheet.mjs (was
  module-private; one-line change, function body untouched, plus a doc-comment
  note on why it's exported now).
• Adds a per-subject assertion in
  scripts/__tests__/plugin-published-stylesheet.test.ts that the emitted
  sheet opens with the header the subject is actually configured to use —
  its own declared buildOptions.header for fields, else the shared
  defaultHeader(PACKAGE_NAME) this module now exports for plugin-grid /
  plugin-kanban — never a second, hand-spelled copy of either wording (the
  route ruled out on the issue as the "assertion that inspects nothing"
  shape).

Why two checks, not one

The obvious single assertion — recompute "expected" by calling the very
function (defaultHeader, or buildOptions.header) that also produced
"actual" — is vacuous: since both sides call the same pure function on the
same input, no mutation to that function's own wording can ever make it
disagree with itself. Ablating confirmed this (see Testing below), so the
assertion adds two independent checks that don't have that shape:

1. Assembly check (all 3 subjects): the emitted sheet must start with
   expectedHeader, where expectedHeader is read from the subject's own
   build script export. Catches the header not being prepended at all, or
   some unrelated value being used.
2. Name-presence check (all 3 subjects): the banner region contains the
   subject's own PACKAGE_NAME — a plain string constant each build script
   declares independently, never derived from defaultHeader. This is what
   gives check 1 teeth against a broken interpolation inside defaultHeader
   itself, for the two packages that inherit it.
3. fields-only control: fields' emitted sheet must NOT start with the
   shared default banner. This is the check that actually catches "an
   accidental drop of fields' header override" — the scenario named in the
   issue — because dropping it moves fields' "expected" (check 1's fallback)
   and "actual" to the shared default in lockstep, so check 1 alone can't see
   it.

Testing

pnpm exec vitest run scripts/__tests__/plugin-published-stylesheet.test.ts --maxWorkers=2
# Test Files 1 passed (1) / Tests 24 passed (24)
pnpm type-check:scripts # clean
pnpm check:control-bytes # OK, 6004 tracked text files scanned
pnpm check:entry-guard # 52 export bindings, 52 inert on import (0 known-unsafe)
pnpm check:esm-specifiers # unaffected (no ESM specifier surface touched)
node scripts/check-changeset-presence.mjs # no changeset owed (scripts/-only diff)

Ablation 1 — mutated defaultHeader() in
scripts/build-plugin-stylesheet.mjs to hardcode a wrong package name
(breaking its package-name interpolation), confirmed the mutation landed on
disk (git diff, byte count), reran the suite:

FAIL plugin-grid > opens the emitted sheet with the banner it declares
FAIL plugin-kanban > opens the emitted sheet with the banner it declares
PASS fields > opens the emitted sheet with the banner it declares
PASS fields > does not fall back to the shared default banner
Tests 2 failed | 22 passed (24)

Restored the mutated file from HEAD (git checkout HEAD on that one path);
git hash-object on the restored file matched the pre-mutation blob exactly.

Ablation 2 — mutated packages/fields/scripts/build-css.mjs to drop its
header: HEADER, line from buildOptions (simulating the exact "accidental
drop" defect named in the issue), confirmed on disk, reran the suite:

PASS plugin-grid > opens the emitted sheet with the banner it declares
PASS plugin-kanban > opens the emitted sheet with the banner it declares
PASS fields > opens the emitted sheet with the banner it declares (self-consistent — expected)
FAIL fields > does not fall back to the shared default banner
Tests 1 failed | 23 passed (24)

Restored the mutated file from HEAD the same way; git hash-object matched
the pre-mutation blob exactly.

Byte-identity of the published artifact (objectui#6405's acceptance
gate): built all three subjects' emitted sheets against this branch's
scripts/build-plugin-stylesheet.mjs and, separately, against the pre-PR
version at 64d624ded (temporarily checked out, restored by blob hash
afterward). SHA-256 of all three emitted sheets is identical before/after —
the export keyword and doc-comment addition change zero published bytes.

Scope

Touches only scripts/build-plugin-stylesheet.mjs (the one-line export +
a doc-comment) and scripts/__tests__/plugin-published-stylesheet.test.ts.
The three build-css.mjs consumers are untouched in the committed diff (they
were only mutated transiently, on disk, during the two ablations above, and
restored by verified blob hash — never part of any commit).

No changeset: scripts-only diff, nothing published moves.

🤖 Generated with Claude Code

https://claude.ai/code/session_01BGMDbrVa8JjZcCQ7DWYH1b