@hachej/boring-agent 0.4.2 → 0.4.3

package/dist/file-types.d.ts

@@ -1,13 +1,15 @@
+import { SNIFF_BYTES } from '@hachej/boring-files/file-types';
 export type FileKind = 'text' | 'pdf' | 'office' | 'image' | 'archive' | 'binary';
 /** The media type of a native image (png, jpeg, gif, webp) by extension, or undefined. */
 export declare const nativeImageType: (path: string) => string | undefined;
-/** How many leading bytes `fileKind` looks at. */
-export declare const SNIFF_BYTES = 16;
+/** How many leading bytes `fileKind` looks at: the file-types table's own. */
+export { SNIFF_BYTES };
 /** The kind of `path`, from its extension and, when given, its leading bytes (the magic number wins). Unknown extensions are text unless a NUL byte says otherwise. */
 export declare function fileKind(path: string, head?: Uint8Array): FileKind;
 /**
  * The media type a converter should be told. A signature in the leading bytes wins over everything (a PDF named `.txt` is
  * `application/pdf`, whatever the provider derived from the name); then the provider's type unless it is generic; then the table's.
+ * Sniffed `text/plain` is not a signature: it never overrides a name.
  */
 export declare function mediaTypeFor(path: string, kind: FileKind, provided: string | undefined, head?: Uint8Array): string;
 //# sourceMappingURL=file-types.d.ts.map

package/dist/file-types.js

@@ -11,7 +11,7 @@
  * A file is classified by its extension AND its leading bytes: a PDF or an image is recognised by its magic number even under a
  * wrong name.
  */
-import { MEDIA_TYPES, extensionOf, sniffMediaType } from '@hachej/boring-files/file-types';
+import { MEDIA_TYPES, SNIFF_BYTES, extensionOf, sniffMediaType } from '@hachej/boring-files/file-types';
 const TEXT = ['md', 'markdown', 'txt', 'csv', 'tsv', 'json', 'jsonl', 'html', 'htm', 'svg', 'xml', 'js', 'mjs', 'cjs', 'ts', 'tsx', 'jsx', 'css', 'py', 'sql', 'yaml', 'yml', 'sh', 'toml', 'ini', 'log', 'rst', 'tex'];
 const OFFICE = ['doc', 'docx', 'odt', 'rtf', 'xls', 'xlsx', 'ods', 'ppt', 'pptx', 'odp', 'epub'];
 /** Image extensions; `native` ones a model reads as an image part without conversion. Their media types are `MEDIA_TYPES`. */
@@ -24,16 +24,24 @@ const BY_EXTENSION = new Map([
 ]);
 /** The media type of a native image (png, jpeg, gif, webp) by extension, or undefined. */
 export const nativeImageType = (path) => IMAGES[extensionOf(path)] === true ? MEDIA_TYPES[extensionOf(path)] : undefined;
-/** How many leading bytes `fileKind` looks at. */
-export const SNIFF_BYTES = 16;
+/** How many leading bytes `fileKind` looks at: the file-types table's own. */
+export { SNIFF_BYTES };
+/** The signatures this file routes as an image or an office document, as `sniffMediaType` names them. */
+const SNIFFED_IMAGES = new Set(['image/png', 'image/jpeg', 'image/gif', 'image/webp']);
+const SNIFFED_OFFICE = new Set(OFFICE.map(extension => MEDIA_TYPES[extension]));
 const startsWith = (bytes, signature, at = 0) => signature.every((byte, index) => bytes[at + index] === byte);
-/** The kind a magic number names, or undefined. A ZIP signature is an office document only under an office extension. */
+/**
+ * The kind a magic number names, or undefined. `sniffMediaType` names the signatures (an office document by its zip entries); a zip
+ * it cannot name is an office document only under an office extension, and an OLE container (legacy doc, xls, ppt) is one.
+ */
 function kindByMagic(bytes, extension) {
     const sniffed = sniffMediaType(bytes);
     if (sniffed === 'application/pdf')
         return 'pdf';
-    if (sniffed?.startsWith('image/') || (startsWith(bytes, [0x42, 0x4d]) && extension === 'bmp'))
+    if ((sniffed !== undefined && SNIFFED_IMAGES.has(sniffed)) || (startsWith(bytes, [0x42, 0x4d]) && extension === 'bmp'))
         return 'image';
+    if (sniffed !== undefined && SNIFFED_OFFICE.has(sniffed))
+        return 'office';
     if (startsWith(bytes, [0x50, 0x4b, 0x03, 0x04]))
         return OFFICE.includes(extension) ? 'office' : 'archive';
     if (startsWith(bytes, [0xd0, 0xcf, 0x11, 0xe0]))
@@ -56,10 +64,11 @@ export function fileKind(path, head) {
 /**
  * The media type a converter should be told. A signature in the leading bytes wins over everything (a PDF named `.txt` is
  * `application/pdf`, whatever the provider derived from the name); then the provider's type unless it is generic; then the table's.
+ * Sniffed `text/plain` is not a signature: it never overrides a name.
  */
 export function mediaTypeFor(path, kind, provided, head) {
     const sniffed = head === undefined ? undefined : sniffMediaType(head);
-    if (sniffed !== undefined)
+    if (sniffed !== undefined && sniffed !== 'text/plain')
         return sniffed;
     const extension = extensionOf(path);
     const table = kind === 'pdf' ? MEDIA_TYPES.pdf : kind === 'image' && IMAGES[extension] !== undefined ? MEDIA_TYPES[extension] : kind === 'office' && OFFICE.includes(extension) ? MEDIA_TYPES[extension] : undefined;

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@hachej/boring-agent",
-  "version": "0.4.2",
+  "version": "0.4.3",
   "license": "MIT",
   "private": false,
   "type": "module",
@@ -143,8 +143,8 @@
     "@earendil-works/pi-durable": "~1.2.0",
     "@earendil-works/pi-ai": "~1.2.0",
     "@earendil-works/chord": "~1.2.0",
-    "@hachej/boring-files": "^0.4.2",
-    "@hachej/boring-ui-kit": "^0.4.2"
+    "@hachej/boring-files": "^0.4.3",
+    "@hachej/boring-ui-kit": "^0.4.3"
   },
   "peerDependenciesMeta": {
     "@hachej/boring-files": {