@shbernal/ts-xlsx 3.2.1 → 3.2.2-next.22.g076b1ec

package/README.md

@@ -7,25 +7,25 @@
 [![License: MIT](https://img.shields.io/badge/license-MIT-blue.svg)](LICENSE)
 
 A TypeScript-first library for reading and writing spreadsheet documents
-(`.xlsx` / OOXML, and CSV) — synchronous, `Uint8Array`-native, and dependency-lean.
+(`.xlsx` / OOXML, and CSV). It is synchronous, `Uint8Array`-native, and dependency-lean.
 
 **[Documentation, guide and a browser playground](https://shbernal.github.io/ts-xlsx/)**
 
-> **Status — 1.0.0, the first release.**
+> **Status: stable and published on npm.** The [changelog](CHANGELOG.md) has the current version.
 > `ts-xlsx` began as a hard fork of [ExcelJS](https://github.com/exceljs/exceljs)
 > and has been rebuilt from the ground up into a modern, strict-TypeScript library.
-> **It carries no backwards-compatibility guarantee with ExcelJS** — the API below is
+> **It carries no backwards-compatibility guarantee with ExcelJS**: the API below is
 > its own, not a drop-in. From 1.0.0 onward it follows
 > [SemVer](https://semver.org/) against _itself_; see the [changelog](CHANGELOG.md).
-> See [`CLAUDE.md`](CLAUDE.md) for the goals and [`docs/architecture.md`](docs/architecture.md)
+> See [`AGENTS.md`](AGENTS.md) for the goals and [`docs/architecture.md`](docs/architecture.md)
 > for the design, and [migrating from ExcelJS](docs/migrating-from-exceljs.md) if you are coming across.
 
 ## Why it exists
 
-Upstream ExcelJS is effectively unmaintained — no release since 2023, a backlog of
-hundreds of issues and PRs — while still serving tens of millions of downloads a month.
+Upstream ExcelJS is effectively unmaintained (no release since 2023, a backlog of
+hundreds of issues and PRs), while still serving tens of millions of downloads a month.
 `ts-xlsx` extracts the accumulated value from that backlog (as a permanent regression
-corpus) and discards the debt: no untyped surfaces, no callback APIs, a single small
+corpus) and discards the debt: no untyped APIs, no callback APIs, a single small
 runtime dependency, and every behavior pinned by a test. This is not a compatibility
 shim. It is a different, better library that happens to share ancestry.
 
@@ -35,9 +35,9 @@ What that buys you today:
   `exactOptionalPropertyTypes`. The published `.d.ts` _is_ the documentation, and the
   [API reference](docs/api/README.md) is generated straight from it.
 - **Synchronous, buffer-native I/O.** `readXlsx`/`writeXlsx` take and return a
-  `Uint8Array` — no streams to await for the buffered path, no Node `Buffer` assumption,
+  `Uint8Array`. There are no streams to await for the buffered path, no Node `Buffer` assumption,
   so the same code runs in Node and the browser.
-- **Streaming when you need it.** Bounded-memory row streaming both ways — read a large
+- **Streaming when you need it.** Bounded-memory row streaming both ways: read a large
   workbook without materializing it, or write one incrementally to a Node stream so peak
   memory stays flat.
 - **One runtime dependency** ([`fflate`](https://github.com/101arrowz/fflate) for zip),
@@ -77,7 +77,7 @@ const people = reopened.getWorksheet('People');
 console.log(people?.getCell('A2').value); // 'Ada'
 ```
 
-Persisting to disk is your platform's job, not the library's — `writeXlsx` hands you the
+Persisting to disk is your platform's job, not the library's. `writeXlsx` hands you the
 bytes:
 
 ```ts
@@ -89,17 +89,17 @@ const wb2 = readXlsx(readFileSync('people.xlsx')); // a Buffer is a Uint8Array
 
 ## Core model
 
-- **`Workbook`** — the document. `addWorksheet(name)`, `getWorksheet(nameOrId)`,
+- **`Workbook`**: the document. `addWorksheet(name)`, `getWorksheet(nameOrId)`,
   `worksheets`, defined names, images, and workbook-level properties.
-- **`Worksheet`** — a sheet. `getCell('B3')`, `addRow(values)`, `addTable(...)`,
+- **`Worksheet`**: a sheet. `getCell('B3')`, `addRow(values)`, `addTable(...)`,
   `mergeCells('A1:B2')`, `getRow(n)` / `getColumn(n)`, `rows()` / `columns()`, page setup,
   and print options.
-- **`Row`** / **`Column`** — one line of the grid. Formatting is flat (`row.height = 20`,
+- **`Row`** / **`Column`**: one line of the grid. Formatting is flat (`row.height = 20`,
   `column.width = 12`, `column.key = 'name'`), and cells are reachable from it:
   `row.getCell('B')`, `row.cells`, `row.values`.
-- **`Cell`** — one cell. `cell.value` is the whole story: a `number`, `string`,
+- **`Cell`**: one cell. `cell.value` is the whole story: a `number`, `string`,
   `boolean`, `Date`, `null` (empty), a formula (`{formula, result}`), rich text, or an
-  error — all precisely typed as [`CellValue`](docs/api/cell-values.md). A hyperlink is not
+  error, all precisely typed as [`CellValue`](docs/api/cell-values.md). A hyperlink is not
   a value: it sits on the sheet, over any cell, through
   `sheet.addHyperlink({ref: 'B2', target: 'https://example.com'})`.
 
@@ -115,16 +115,16 @@ for (const row of sheet.rows()) {
 ```
 
 `Row` and `Column` are _handles_, not snapshots: they read and write straight through to the
-sheet, so two handles on the same line always agree, and reading one creates nothing — asking
+sheet, so two handles on the same line always agree, and reading one creates nothing. Asking
 about row 500 costs nothing and does not extend the used range. Position is fixed, exactly as a
 `Cell`'s is: after a splice, `getRow(3)` still means row 3, now holding whatever moved there.
 
 Addresses are honest: an axis a reference doesn't mention is `undefined`, never a
-sentinel — see [`decodeAddress`](docs/api/addresses-ranges.md).
+sentinel. See [`decodeAddress`](docs/api/addresses-ranges.md).
 
 Charts, vector shapes, slicers, and legacy form controls are **round-trip-only**: a
 workbook that has them keeps them byte-faithfully through a load/edit/save, but there is
-no API to author a new one — see [`docs/api/preserved.md`](docs/api/preserved.md) and
+no API to author a new one. See [`docs/api/preserved.md`](docs/api/preserved.md) and
 [ADR-0014](docs/decisions/0014-charts-shapes-slicers-are-round-trip-only-for-1-0.md).
 
 ## Reading, writing, streaming, CSV
@@ -140,10 +140,10 @@ import {
   writeCsvText, // CSV as a string
 } from '@shbernal/ts-xlsx';
 // The streaming writer opens files and pipes Node streams, so it is the one Node-only face
-// and lives behind its own subpath — which is what keeps everything above browser-safe:
+// and lives behind its own subpath. That keeps everything above browser-safe:
 import {WorkbookStreamWriter} from '@shbernal/ts-xlsx/node';
 
-// Bounded-memory extraction — the whole workbook is never materialized:
+// Bounded-memory extraction. The whole workbook is never materialized:
 for (const row of readSheetRows(bytes, {sheet: 'People'})) {
   console.log(
     row.number,
@@ -151,7 +151,7 @@ for (const row of readSheetRows(bytes, {sheet: 'People'})) {
   );
 }
 
-// Bounded-memory generation — commit each row to serialize and free it as you go:
+// Bounded-memory generation. Commit each row to serialize and free it as you go:
 const writer = new WorkbookStreamWriter();
 const out = writer.addWorksheet('Big');
 for (let i = 1; i <= 1_000_000; i++) out.addRow([i, i * i]).commit();
@@ -168,7 +168,7 @@ never assembled as one object, which is what passing a sink was for.
 
 ## Writing without blocking the event loop
 
-`writeXlsx` spends the whole cost of DEFLATE on the calling thread — seconds, for a large
+`writeXlsx` spends the whole cost of DEFLATE on the calling thread: seconds, for a large
 workbook, during which a server answers nothing. `writeXlsxAsync` produces the same package
 with the compression handed to worker threads:
 
@@ -185,12 +185,12 @@ always and speed only when there is more than one substantial part.
 
 There is deliberately **no `readXlsxAsync`**. Reading is dominated by XML parsing and model
 building, which no worker can take off the calling thread, so it would advertise a
-non-blocking read and then block for most of its duration — and the reader's zip-bomb ceiling
-is enforced by counting output between synchronous input slices, a guarantee that weakens the
+non-blocking read and then block for most of its duration. The reader also enforces its zip-bomb
+ceiling by counting output between synchronous input slices, a guarantee that weakens the
 moment inflation moves to a worker. If you need a non-blocking read, run the whole read in a
 worker. See [ADR-0024](docs/decisions/0024-async-is-one-writer-not-a-mirrored-pair.md).
 
-The reader decodes untrusted input defensively — entities are decoded but never
+The reader decodes untrusted input defensively. Entities are decoded but never
 expanded, and inflation is bounded by a running output counter rather than any declared
 size, so a malformed or hostile package can't exhaust memory.
 
@@ -199,7 +199,7 @@ size, so a malformed or hostile package can't exhaust memory.
 The bare package name gives you everything, and with a bundler that is the right default:
 `"sideEffects": false` is declared, so anything you don't reference is dropped. The subpaths
 are for when you'd rather the module graph itself said which half of the library you depend
-on — a Lambda with no bundler, a service that only classifies failures:
+on, as in a Lambda with no bundler or a service that only classifies failures:
 
 | Import from                  | You get                                                                        | It loads |
 | ---------------------------- | ------------------------------------------------------------------------------ | -------- |
@@ -219,18 +219,18 @@ no Node built-in at all, which is what makes them bundle for a browser with noth
 configure ([ADR-0040](docs/decisions/0040-the-browser-boundary-is-an-entry-point.md)).
 
 Every error class lives in `/errors` and nowhere else, because a container-level failure
-belongs to no single codec — `readXlsx` and `readXlsb` both raise `UnsupportedFormatError`.
+belongs to no single codec: `readXlsx` and `readXlsb` both raise `UnsupportedFormatError`.
 Catching and classifying therefore costs 3 KB, not a parser.
 
 `/xlsx` is barely cheaper than the whole package, and that is honest rather than a defect:
 `readXlsx` sniffs the bytes and hands a binary package to the BIFF12 reader, so the `.xlsb`
 codec is not optional on that path. The numbers above are the static-import closure of each
-entry, measured by `pnpm run size` — a bundler can only shrink them further.
+entry, measured by `pnpm run size`. A bundler can only shrink them further.
 
 ## API reference
 
-The full reference is generated from the public types — it cannot drift from what the
-compiler accepts — and lives in **[`docs/api/`](docs/api/README.md)**. Regenerate it with:
+The full reference is generated from the public types, so it cannot drift from what the
+compiler accepts. It lives in **[`docs/api/`](docs/api/README.md)**. Regenerate it with:
 
 ```shell
 pnpm run docs
@@ -246,7 +246,7 @@ shapes are deliberately different (synchronous `Uint8Array` I/O instead of async
 ## Hit a bug? There is a skill for that
 
 Most code that uses this library is written by an agent, and an agent that hits a library
-defect will usually route around it silently — so the defect is never reported and never
+defect will usually route around it silently, so the defect is never reported and never
 fixed. `ts-xlsx-upstream` is a skill that turns that moment into a filed issue with a
 minimal reproduction, which is what the maintainers convert into a permanent regression
 test. It ships inside the package, so it is already on disk:
@@ -262,8 +262,8 @@ copies: they are derived, `npx skills experimental_install` rebuilds them from t
 `npx skills update` overwrites them in place.
 
 It covers triage (is this bug ours or your file's?), reducing a failure to a script that
-builds its own input, and — because spreadsheets carry real data — never uploading a
-workbook to a public tracker. Once the reproduction stands on its own, it files without
+builds its own input, and never uploading a workbook to a public tracker, because
+spreadsheets carry real data. Once the reproduction stands on its own, it files without
 interrupting you, and tells you the issue number afterwards.
 
 You do not need it to report something: <https://github.com/shbernal/ts-xlsx/issues> is
@@ -273,10 +273,10 @@ open, and errors the library knows are its own fault print that link themselves.
 
 This project is optimized to be built and maintained largely by autonomous agents, with
 a machine-checkable safety net as the primary guarantor of correctness. The rules that
-govern every change are in [`CLAUDE.md`](CLAUDE.md); the design and working agreements are
+govern every change are in [`AGENTS.md`](AGENTS.md); the design and working agreements are
 in [`docs/architecture.md`](docs/architecture.md); notable decisions are recorded under
 [`docs/decisions/`](docs/decisions/).
 
 ## License
 
-[MIT](LICENSE) — inherited from ExcelJS and retained.
+[MIT](LICENSE), inherited from ExcelJS and retained.

package/dist/core/autofilter.d.ts

@@ -50,7 +50,7 @@ export declare const isCustomFilterOperator: (value: string) => value is CustomF
  * is the common case (dropdowns, no criteria); an {@link AutoFilter} object carries per-column
  * criteria too. Throws when the range is not a bounded rectangle, or a column's `colId` falls
  * outside it, or a custom filter does not carry one or two predicates: a filter authored wrong is
- * a bug to surface, not to silently repair. (The reader sanitises hostile input before it reaches
+ * a bug to report, not to silently repair. (The reader sanitises hostile input before it reaches
  * here, so load never trips these guards.)
  */
 export declare function canonicalizeAutoFilter(input: string | AutoFilter): AutoFilter;

package/dist/core/data-validation-overlay.d.ts

@@ -14,7 +14,7 @@ export declare class DataValidationOverlay {
      * round-trip writes it back there instead of silently corrupting the cross-sheet reference.
      *
      * @throws {AuthoringError} when `sqref` names no area at all. A rule attached to nothing covers no
-     * cell and is written back as the same unreadable text, so it is a mistake worth surfacing at the
+     * cell and is written back as the same unreadable text, so it is a mistake worth reporting at the
      * call. The reader does not reach this: it drops such an entry at its own boundary, where a foreign
      * file's malformed attribute is supposed to be dropped.
      */

package/dist/core/workbook.d.ts

@@ -14,17 +14,21 @@ import { type NamedCellStyle, type TableStyleTable } from './workbook-styles.ts'
 import { type DeclaredThemeSchemes } from './workbook-theme.ts';
 import { type Visibility, Worksheet, type WorksheetState } from './worksheet.ts';
 /**
- * A workbook-level reference to package content the model does not model: a pivot cache
- * (`pivotCacheDefinition`) or a slicer cache (`slicerCache`), preserved verbatim across a round-trip
- * instead of being dropped. `relType` is the workbook relationship Type URI to re-emit; `entryPath`
- * is the part it points at; `parts` is the transitive closure that reference reaches (the entry
- * included). `pivotCacheId` carries the `<pivotCache cacheId>` a pivot cache is registered under in
- * the workbook's `<pivotCaches>`, so the wiring a pivot table resolves its cache through is re-emitted
- * too; it is absent for a slicer cache, which the workbook lists in an extension block instead.
+ * A workbook-level reference to package content the model does not interpret: a pivot cache
+ * (`pivotCacheDefinition`), a slicer cache (`slicerCache`) or an external link. The reader keeps it
+ * verbatim so a round-trip re-emits it instead of dropping it.
+ *
+ * `relType` is the workbook relationship Type URI to re-emit, `entryPath` is the part it points at,
+ * and `parts` is the transitive closure that reference reaches, the entry included.
+ *
+ * `pivotCacheId` is the `<pivotCache cacheId>` a pivot cache is registered under in the workbook's
+ * `<pivotCaches>`. A pivot table finds its cache through that id, so the writer re-emits it too. A
+ * slicer cache has none, because the workbook lists slicer caches in an extension block instead.
+ *
  * `externalReferenceIndex` is the 0-based position of an `externalLink` within the workbook's
- * `<externalReferences>`, the `[n]` a formula or defined name resolves an external cell through, so
- * the block is re-emitted in the original order and every `[n]` still points at the same linked
- * workbook; it is absent for a pivot/slicer cache.
+ * `<externalReferences>`: the `[n]` through which a formula or defined name reaches an external
+ * cell. The writer re-emits the block in its original order, so every `[n]` still points at the same
+ * linked workbook. A pivot or slicer cache has none.
  */
 export interface PreservedWorkbookReference {
     readonly relType: string;
@@ -113,7 +117,7 @@ export interface WorkbookProperties {
     company?: string;
 }
 /**
- * A named reference in the workbook, one of the entries Excel surfaces in its Name Manager. A name
+ * A named reference in the workbook, one of the entries Excel lists in its Name Manager. A name
  * maps an identifier to a formula (`refersTo`), most often a cell range like `Sheet1!$A$1:$B$2` but
  * possibly any formula. A name is global to the workbook unless it names a sheet in {@link scope},
  * which restricts it to that sheet and lets another sheet reuse the same name independently.
@@ -369,8 +373,8 @@ export declare class Workbook {
      * read from a package carrying no styles part: nothing was declared, and the library does not
      * fabricate a declaration on the file's behalf.
      *
-     * This is the *round-trip* surface. {@link defaultFont} is what the workbook actually renders in,
-     * which is this once anything has been authored over it.
+     * This is the value a round-trip preserves. {@link defaultFont} is what the workbook actually
+     * renders in, which is this once anything has been authored over it.
      */
     get declaredDefaultFont(): Font | undefined;
     /**
@@ -496,14 +500,14 @@ export declare class Workbook {
     /**
      * Show `images` on `sheet`, a worksheet of this workbook, registering each picture's bytes here
      * and re-anchoring it against the id they land on. The counterpart to {@link exportImages}, and
-     * the affordance that lets a picture cross workbooks at all: an {@link AnchoredImage}'s `imageId`
-     * indexes one workbook's registry and means nothing in the next, so a raw anchor moved between
-     * workbooks points at media that does not exist there.
+     * the only way a picture can cross workbooks: an {@link AnchoredImage}'s `imageId` indexes one
+     * workbook's registry and means nothing in the next, so a raw anchor moved between workbooks
+     * points at media that does not exist there.
      *
      * The sheet's existing pictures are replaced, not appended to, so `importImages` is a transfer
-     * rather than an accumulation: the same direction {@link Worksheet.model} assignment goes, and
-     * what makes re-importing a sheet's own export leave it unchanged. An import whose `background` is
-     * absent clears the destination's background for the same reason.
+     * rather than an accumulation. {@link Worksheet.model} assignment works the same way, and it is why
+     * re-importing a sheet's own export leaves the sheet unchanged. For the same reason, an import with
+     * no `background` clears the destination's background.
      *
      * Registration is content-addressed: a picture whose bytes are already here is re-used at its
      * existing id rather than stored twice, so importing the same sheet repeatedly, or twenty sheets
@@ -532,7 +536,7 @@ export declare class Workbook {
      * {@link getWorksheet}, for a caller who knows the sheet is there: the miss throws instead of
      * returning `undefined`, and the message names every sheet the workbook does have.
      *
-     * The partial lookup is the right primitive for asking *whether* a sheet exists, and the wrong
+     * The partial lookup is the right call for asking *whether* a sheet exists, and the wrong
      * one for reaching a sheet a template is expected to carry: `undefined` flows on into a `?.`
      * chain and fails several steps later with nothing left to say about which name was missing.
      * That listing is the whole point: a lookup miss is a typo, a stale template or a renamed tab,

package/dist/core/worksheet.d.ts

@@ -342,7 +342,7 @@ export declare class Worksheet {
     /**
      * The rows to serialise, as handles, in ascending row order: the union of rows holding cells and
      * rows holding only metadata (a hidden or grouped row need carry no data). Mirrors how OOXML
-     * serialises (`<row>` wrapping `<c>`) and is the writer's row surface.
+     * serialises (`<row>` wrapping `<c>`) and is what the writer iterates over.
      *
      * A handle yields its cells only when asked, so a pass that reads nothing but row attributes
      * never assembles a cell array it will not look at.
@@ -453,7 +453,7 @@ export declare class Worksheet {
     }, properties?: PictureProperties): void;
     /**
      * Anchor an image with a pre-built model anchor in the model's own units (EMUs). This is the
-     * low-level primitive {@link addImage} builds on and the reader uses to re-pin an image parsed from
+     * low-level call {@link addImage} builds on and the reader uses to re-pin an image parsed from
      * a drawing part without a lossy pixel round-trip.
      */
     addImageAnchor(imageId: number, anchor: ImageAnchor, properties?: PictureProperties): void;

package/dist/customui/ribbon.d.ts

@@ -11,7 +11,7 @@ export declare const CUSTOMUI_2010_REL_TYPE = "http://schemas.microsoft.com/offi
 export declare function isCustomUiRelType(type: string): boolean;
 /**
  * A control element inside a ribbon group. `kind` is the element's local name, narrowed to the closed
- * set of RibbonX control elements ({@link RibbonControlKind}); an element outside that set is surfaced
+ * set of RibbonX control elements ({@link RibbonControlKind}); an element outside that set is kept
  * as `unknown` rather than dropped. The three identity attributes (`id` a document-defined control,
  * `idQ` a qualified id, `idMso` a built-in control) and the two most-consulted display/behaviour
  * attributes (`label`, `onAction`) are lifted out as typed conveniences; every attribute the element

package/dist/io/csv/write.d.ts

@@ -5,8 +5,8 @@ import type { Workbook } from '../../core/workbook.ts';
  * meaning the same bytes.
  *
  * Named here rather than taken from Node's `BufferEncoding`, which is what this option used to be:
- * that type is part of `@types/node`, so it made a browser consumer's public API surface depend on
- * Node's types, and it offered `base64` and `hex` as if they were output encodings for a text
+ * that type is part of `@types/node`, so it made a browser consumer's public API depend on Node's
+ * types, and it offered `base64` and `hex` as if they were output encodings for a text
  * format. The list is what a CSV consumer actually asks for.
  */
 export type CsvEncoding = 'ascii' | 'latin1' | 'ucs-2' | 'ucs2' | 'utf-8' | 'utf-16le' | 'utf16le' | 'utf8';

package/dist/io/style/xf-style.d.ts

@@ -28,7 +28,7 @@ export interface StyleTable {
     readonly namedStyles: ReadonlyArray<NamedCellStyle>;
     /**
      * Font id 0: the workbook's declared default font, the face every cell naming no font renders in.
-     * Surfaced separately from the fonts it was flattened onto because it is workbook-level state, not a
+     * Kept separately from the fonts it was flattened onto because it is workbook-level state, not a
      * cell format: a re-write must emit *this* face as font 0 rather than an assumed Calibri, or every
      * empty cell changes face and every character-unit column width changes meaning. Absent when the
      * file declares no font table.

package/dist/io/xlsx/comments.d.ts

@@ -38,7 +38,7 @@ export declare function collectNotes(cells: Iterable<Cell>): CommentCell[];
  *
  * Authors are laid out the way Excel lays them out: one synthetic `tc={headId}` entry per threaded
  * conversation first, then a single anonymous author shared by every note (the model carries no note
- * author). Each comment points at its own author by index, and a fallback additionally carries the
+ * author). Each comment points at its own author by index, and a fallback also carries the
  * `xr:uid` naming its thread: the pair that keeps Excel treating the cell as threaded.
  */
 export declare function commentsXml(comments: readonly CommentCell[]): string;

package/dist/vba/project-editor.d.ts

@@ -7,7 +7,7 @@
  * Only `procedural` and `class` modules can be removed this way. Removing a `document` module (e.g.
  * `ThisWorkbook`) or a `designer` module (a UserForm) would leave the host referencing code that no
  * longer exists, since their names are tied to a worksheet/workbook `codeName` or a designer storage
- * this project-level primitive has no visibility into. Editing such a module's code-behind is a job for
+ * this project-level edit has no visibility into. Editing such a module's code-behind is a job for
  * the offline `tools/vba-compiler` (in-place mode), which drives the real host.
  *
  * @throws {VbaParseError} if `bin` is not a parseable VBA project (validated before any edit).

package/dist/xml/xml-read.d.ts

@@ -122,7 +122,7 @@ export declare function tagRanges(source: string): Generator<TagRange>;
  * {@link XmlParseError}.
  */
 export declare function elementRanges(source: string, local: string): Generator<ElementRange>;
-/** An element start surfaced by {@link openElements}: its qualified `name`, the namespace-stripped
+/** An element start yielded by {@link openElements}: its qualified `name`, the namespace-stripped
  * `local` name the filter matched on, and its already-decoded `attrs`. */
 export interface OpenElement {
     readonly name: string;

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@shbernal/ts-xlsx",
-  "version": "3.2.1",
+  "version": "3.2.2-next.22.g076b1ec",
   "description": "A TypeScript-first library for reading and writing xlsx (OOXML) spreadsheets.",
   "keywords": [
     "csv",
@@ -88,7 +88,6 @@
     "site:preview": "vitepress preview www",
     "docs": "node scripts/gen-docs.ts",
     "docs:check": "node scripts/gen-docs.ts && git add --intent-to-add -- docs/api && git diff --exit-code -- docs/api",
-    "constitution:check": "node scripts/check-constitution.ts",
     "layering:check": "node scripts/check-layering.ts",
     "corpus-blind:check": "node scripts/check-corpus-blind.ts",
     "entries:check": "node scripts/check-entries.ts",
@@ -125,18 +124,18 @@
     "fflate": "^0.8.3"
   },
   "devDependencies": {
-    "@types/node": "^24.13.3",
+    "@types/node": "^24.19.2",
     "charcheck": "^0.4.0",
-    "jszip": "^3.10.1",
-    "lefthook": "^2.1.12",
+    "jszip": "^3.10.2",
+    "lefthook": "^2.2.1",
     "ooxml-validate": "^0.0.3",
-    "oxfmt": "^0.67.0",
-    "oxlint": "^1.82.0",
-    "oxlint-tsgolint": "7.0.2001",
-    "shiki": "^4.4.3",
+    "oxfmt": "^0.72.0",
+    "oxlint": "^1.87.0",
+    "oxlint-tsgolint": "7.0.2003",
+    "shiki": "^4.5.0",
     "typescript": "^7.0.2",
     "vitepress": "^1.6.4",
-    "vue": "^3.5.42"
+    "vue": "^3.5.43"
   },
   "engines": {
     "node": ">=24"

package/skills/ts-xlsx-upstream/SKILL.md

@@ -17,7 +17,7 @@ builds its own input.** Everything below is in service of that.
 
 ## 1. Decide whether it is actually ours
 
-Catch the error and read `code` — the taxonomy is deliberately coarse and each answer
+Catch the error and read `code`. The taxonomy is deliberately coarse, and each answer
 means something different about whose bug it is.
 
 | `error.code`         | Whose bug        | Report it?                                                      |
@@ -45,14 +45,14 @@ node -v                                                       # runtime
 ```
 
 Also capture, verbatim: the error `name`, its `code`, the full message, and the stack.
-Do not paraphrase the message — the throw site is often identifiable from its exact text.
+Do not paraphrase the message: the throw site is often identifiable from its exact text.
 
 ## 3. Build a minimal reproduction
 
 **Never attach or paste the user's workbook.** Spreadsheets in a real project carry
 salaries, customer lists, and unreleased numbers. Treat every file in this repo as
 confidential unless the user tells you otherwise, and never upload one to a public
-tracker — including "just a screenshot of the sheet".
+tracker, including "just a screenshot of the sheet".
 
 Instead, write a self-contained script that _constructs_ its input and fails:
 
@@ -80,7 +80,7 @@ from that description; they can never unsee an attachment.
 
 If you need a file attached, build one: reproduce the _structure_ you suspect in a workbook
 you generate, with invented values. A synthesized file is always safe to attach. A redacted
-one is not yours to judge — redaction fails quietly, and a public tracker is permanent.
+one is not yours to judge: redaction fails quietly, and a public tracker is permanent.
 
 ## 4. Check it is not already filed
 
@@ -167,7 +167,7 @@ this project and leave a comment pointing at the issue, so the stopgap is remova
 the fix ships:
 
 ```ts
-// Workaround for ts-xlsx#<N> — remove once fixed upstream.
+// Workaround for ts-xlsx#<N>: remove once fixed upstream.
 ```
 
 ## If `gh` is unavailable