@bevel-software/platform-core-backend 0.11.1 → 0.12.0

package/dist/modules/workspace/file-readers/doc-extract.service.js

@@ -0,0 +1,90 @@
+import { extractionMarker, fileExtension } from './doc-extract.types.js';
+import { DocExtractionCache, gitBlobSha } from './extraction-cache.js';
+/**
+ * The content-hash CACHE wrap around document text extraction. The per-format
+ * `DocumentReader`s in the file-reader registry stay pure (each pairs one
+ * extension with one extract function); this service adds the caching
+ * generically — a reader hands its extract function in, and the service only
+ * runs it on a cache miss.
+ *
+ * Caching: keyed by the git BLOB sha of the bytes (see `gitBlobSha`) PLUS the
+ * path's extension, stored as `{ summary, text }` WITHOUT the path — the
+ * marker is assembled per call so the same content read under two paths
+ * (copies, branches) shares one entry yet each read's marker names the path
+ * that was read. The extension is part of the key because identical bytes
+ * extract DIFFERENTLY per format: a `.odt` renamed `.ods` must run the ods
+ * extractor, not return the odt extraction. Extraction is deterministic
+ * (stable slide/sheet/page ordering), which is what makes a content-keyed
+ * cache correct. Failures are NOT cached (rare, and fail fast). The key also carries
+ * `EXTRACTION_SCHEMA`, so an upgrade that changes what an extractor emits does
+ * not keep serving the previous release's text for unchanged bytes.
+ */
+/**
+ * Bumped whenever extraction OUTPUT changes — a new extractor, a fixed one, a
+ * reworded summary, a different marker.
+ *
+ * The rest of the key is content, and content-addressing is only correct while
+ * the same bytes mean the same text. An upgrade that changes what an extractor
+ * emits breaks that: every already-cached document would keep serving the OLD
+ * extraction forever, because its bytes never changed. Bumping this retires
+ * those entries (the cache evicts by age, so they cost nothing for long).
+ *
+ * v2: extraction moved from hand-rolled scanning to a real XML/HTML parser,
+ * which changed self-closing paragraphs, entity edge cases and recovery from
+ * malformed parts.
+ */
+export const EXTRACTION_SCHEMA = 'v2';
+export class DocExtractService {
+    cache;
+    /**
+     * Cold extractions currently running, by cache key. Two reads of the same
+     * document arriving together would otherwise BOTH parse it — the expensive
+     * half of this service, run twice for one answer — because neither had
+     * written the cache entry yet when the other looked.
+     */
+    inFlight = new Map();
+    constructor(cacheRoot) {
+        this.cache = new DocExtractionCache(cacheRoot);
+    }
+    /**
+     * The CACHED extraction for these bytes, or undefined on a cache miss —
+     * never extracts. `grep` uses this (via `DocumentReader.greppableText`) to
+     * search already-extracted documents for free and apply its per-walk budget
+     * only to cold ones.
+     */
+    async getCached(path, bytes) {
+        const hit = await this.cache.get(this.cacheKey(path, bytes));
+        return hit && { marker: extractionMarker(path, hit.summary), text: hit.text };
+    }
+    /** Extract via `extractFn` (cache hit or parse + store). `path` supplies the marker AND the key's format part. */
+    async extract(path, bytes, extractFn) {
+        const key = this.cacheKey(path, bytes);
+        const hit = await this.cache.get(key);
+        if (hit)
+            return { ok: true, marker: extractionMarker(path, hit.summary), text: hit.text };
+        const running = this.inFlight.get(key);
+        const result = await (running ??
+            (() => {
+                // The cache write stays INSIDE the shared promise: were the entry
+                // dropped as soon as parsing settled, a read arriving during the
+                // write would miss both the cache and the in-flight map — and parse
+                // the same document again.
+                const started = (async () => {
+                    const res = await extractFn(bytes);
+                    if (res.ok)
+                        await this.cache.put(key, { summary: res.summary, text: res.text });
+                    return res;
+                })().finally(() => this.inFlight.delete(key));
+                this.inFlight.set(key, started);
+                return started;
+            })());
+        if (!result.ok)
+            return result;
+        return { ok: true, marker: extractionMarker(path, result.summary), text: result.text };
+    }
+    /** Content hash + lowercased extension — e.g. `…sha….odt` (see the class doc). */
+    cacheKey(path, bytes) {
+        return `${gitBlobSha(bytes)}${fileExtension(path)}.${EXTRACTION_SCHEMA}`;
+    }
+}
+//# sourceMappingURL=doc-extract.service.js.map

package/dist/modules/workspace/file-readers/doc-extract.service.js.map

@@ -0,0 +1 @@
+{"version":3,"file":"doc-extract.service.js","sourceRoot":"","sources":["../../../../src/modules/workspace/file-readers/doc-extract.service.ts"],"names":[],"mappings":"AAAA,OAAO,EAAE,gBAAgB,EAAE,aAAa,EAAsC,MAAM,wBAAwB,CAAC;AAC7G,OAAO,EAAE,kBAAkB,EAAE,UAAU,EAAE,MAAM,uBAAuB,CAAC;AAcvE;;;;;;;;;;;;;;;;;;GAkBG;AACH;;;;;;;;;;;;;GAaG;AACH,MAAM,CAAC,MAAM,iBAAiB,GAAG,IAAI,CAAC;AAEtC,MAAM,OAAO,iBAAiB;IACX,KAAK,CAAqB;IAC3C;;;;;OAKG;IACc,QAAQ,GAAG,IAAI,GAAG,EAAkC,CAAC;IAEtE,YAAY,SAAiB;QAC3B,IAAI,CAAC,KAAK,GAAG,IAAI,kBAAkB,CAAC,SAAS,CAAC,CAAC;IACjD,CAAC;IAED;;;;;OAKG;IACH,KAAK,CAAC,SAAS,CAAC,IAAY,EAAE,KAAa;QACzC,MAAM,GAAG,GAAG,MAAM,IAAI,CAAC,KAAK,CAAC,GAAG,CAAC,IAAI,CAAC,QAAQ,CAAC,IAAI,EAAE,KAAK,CAAC,CAAC,CAAC;QAC7D,OAAO,GAAG,IAAI,EAAE,MAAM,EAAE,gBAAgB,CAAC,IAAI,EAAE,GAAG,CAAC,OAAO,CAAC,EAAE,IAAI,EAAE,GAAG,CAAC,IAAI,EAAE,CAAC;IAChF,CAAC;IAED,kHAAkH;IAClH,KAAK,CAAC,OAAO,CAAC,IAAY,EAAE,KAAa,EAAE,SAAoB;QAC7D,MAAM,GAAG,GAAG,IAAI,CAAC,QAAQ,CAAC,IAAI,EAAE,KAAK,CAAC,CAAC;QACvC,MAAM,GAAG,GAAG,MAAM,IAAI,CAAC,KAAK,CAAC,GAAG,CAAC,GAAG,CAAC,CAAC;QACtC,IAAI,GAAG;YAAE,OAAO,EAAE,EAAE,EAAE,IAAI,EAAE,MAAM,EAAE,gBAAgB,CAAC,IAAI,EAAE,GAAG,CAAC,OAAO,CAAC,EAAE,IAAI,EAAE,GAAG,CAAC,IAAI,EAAE,CAAC;QAC1F,MAAM,OAAO,GAAG,IAAI,CAAC,QAAQ,CAAC,GAAG,CAAC,GAAG,CAAC,CAAC;QACvC,MAAM,MAAM,GAAG,MAAM,CAAC,OAAO;YAC3B,CAAC,GAAG,EAAE;gBACJ,kEAAkE;gBAClE,iEAAiE;gBACjE,oEAAoE;gBACpE,2BAA2B;gBAC3B,MAAM,OAAO,GAAG,CAAC,KAAK,IAA4B,EAAE;oBAClD,MAAM,GAAG,GAAG,MAAM,SAAS,CAAC,KAAK,CAAC,CAAC;oBACnC,IAAI,GAAG,CAAC,EAAE;wBAAE,MAAM,IAAI,CAAC,KAAK,CAAC,GAAG,CAAC,GAAG,EAAE,EAAE,OAAO,EAAE,GAAG,CAAC,OAAO,EAAE,IAAI,EAAE,GAAG,CAAC,IAAI,EAAE,CAAC,CAAC;oBAChF,OAAO,GAAG,CAAC;gBACb,CAAC,CAAC,EAAE,CAAC,OAAO,CAAC,GAAG,EAAE,CAAC,IAAI,CAAC,QAAQ,CAAC,MAAM,CAAC,GAAG,CAAC,CAAC,CAAC;gBAC9C,IAAI,CAAC,QAAQ,CAAC,GAAG,CAAC,GAAG,EAAE,OAAO,CAAC,CAAC;gBAChC,OAAO,OAAO,CAAC;YACjB,CAAC,CAAC,EAAE,CAAC,CAAC;QACR,IAAI,CAAC,MAAM,CAAC,EAAE;YAAE,OAAO,MAAM,CAAC;QAC9B,OAAO,EAAE,EAAE,EAAE,IAAI,EAAE,MAAM,EAAE,gBAAgB,CAAC,IAAI,EAAE,MAAM,CAAC,OAAO,CAAC,EAAE,IAAI,EAAE,MAAM,CAAC,IAAI,EAAE,CAAC;IACzF,CAAC;IAED,kFAAkF;IAC1E,QAAQ,CAAC,IAAY,EAAE,KAAa;QAC1C,OAAO,GAAG,UAAU,CAAC,KAAK,CAAC,GAAG,aAAa,CAAC,IAAI,CAAC,IAAI,iBAAiB,EAAE,CAAC;IAC3E,CAAC;CACF"}

package/dist/modules/workspace/file-readers/doc-extract.types.d.ts

@@ -0,0 +1,55 @@
+/**
+ * Server-side document text extraction — shared types.
+ *
+ * The extractors turn an office document (`.docx` / `.pptx` / `.xlsx`), an
+ * OpenDocument file (`.odt` / `.odp` / `.ods`) or a PDF into plain text an
+ * agent can read and grep. They are HONEST about being
+ * lossy: every successful extraction carries a one-line `summary` the consumer
+ * turns into a marker header (`[extracted text of <path> — <summary>]`) so the
+ * reader knows it is looking at extracted text, not the file's bytes.
+ *
+ * The `summary` (not a full marker) is what extractors return and what the
+ * cache stores, because the cache is keyed by CONTENT (git blob sha): the same
+ * document at two workspace paths shares one cache entry, and baking a path
+ * into the cached value would surface the wrong path on the second read.
+ */
+/** A successful extraction: the marker-summary line + the extracted text. */
+export interface ExtractedDoc {
+    /**
+     * One-line description for the marker header, e.g.
+     * `14 slides + notes; layout, images and formatting omitted`.
+     */
+    summary: string;
+    /** The extracted text: paragraphs/rows as lines, with `[slide N]` / `[sheet: Name]` / `[page N]` structure markers. */
+    text: string;
+}
+/**
+ * Extraction outcome. `ok: false` is a TYPED failure (corrupt/unparseable
+ * file) the consumers turn into an honest message — extraction never throws
+ * for bad file content, so a broken upload cannot 500 a read.
+ */
+export type ExtractResult = ({
+    ok: true;
+} & ExtractedDoc) | {
+    ok: false;
+    message: string;
+};
+/**
+ * A format's PURE extract function — bytes in, `ExtractResult` out (async for
+ * pdf.js). One per supported format (extract-docx.ts and friends); each is
+ * paired with its extension by a `DocumentReader` entry in the file-reader
+ * registry, and cache-wrapped by `DocExtractService`.
+ */
+export type ExtractFn = (bytes: Buffer) => ExtractResult | Promise<ExtractResult>;
+/**
+ * Build the honest ONE-LINE header consumers prepend to extracted text.
+ *
+ * One line is a promise the rest of the read path keeps: grep counts on the
+ * marker occupying exactly one, so read_file and grep agree on line numbers.
+ * A path may legally carry a CR or LF, which would forge extra lines and shift
+ * every number after it, so those are shown as escapes rather than obeyed.
+ */
+export declare function extractionMarker(path: string, summary: string): string;
+/** Lowercased extension of `path` including the dot, or '' when there is none. */
+export declare function fileExtension(path: string): string;
+//# sourceMappingURL=doc-extract.types.d.ts.map

package/dist/modules/workspace/file-readers/doc-extract.types.d.ts.map

@@ -0,0 +1 @@
+{"version":3,"file":"doc-extract.types.d.ts","sourceRoot":"","sources":["../../../../src/modules/workspace/file-readers/doc-extract.types.ts"],"names":[],"mappings":"AAAA;;;;;;;;;;;;;;GAcG;AAEH,6EAA6E;AAC7E,MAAM,WAAW,YAAY;IAC3B;;;OAGG;IACH,OAAO,EAAE,MAAM,CAAC;IAChB,uHAAuH;IACvH,IAAI,EAAE,MAAM,CAAC;CACd;AAED;;;;GAIG;AACH,MAAM,MAAM,aAAa,GACrB,CAAC;IAAE,EAAE,EAAE,IAAI,CAAA;CAAE,GAAG,YAAY,CAAC,GAC7B;IAAE,EAAE,EAAE,KAAK,CAAC;IAAC,OAAO,EAAE,MAAM,CAAA;CAAE,CAAC;AAEnC;;;;;GAKG;AACH,MAAM,MAAM,SAAS,GAAG,CAAC,KAAK,EAAE,MAAM,KAAK,aAAa,GAAG,OAAO,CAAC,aAAa,CAAC,CAAC;AAElF;;;;;;;GAOG;AACH,wBAAgB,gBAAgB,CAAC,IAAI,EAAE,MAAM,EAAE,OAAO,EAAE,MAAM,GAAG,MAAM,CAGtE;AAED,kFAAkF;AAClF,wBAAgB,aAAa,CAAC,IAAI,EAAE,MAAM,GAAG,MAAM,CAIlD"}

package/dist/modules/workspace/file-readers/doc-extract.types.js

@@ -0,0 +1,34 @@
+/**
+ * Server-side document text extraction — shared types.
+ *
+ * The extractors turn an office document (`.docx` / `.pptx` / `.xlsx`), an
+ * OpenDocument file (`.odt` / `.odp` / `.ods`) or a PDF into plain text an
+ * agent can read and grep. They are HONEST about being
+ * lossy: every successful extraction carries a one-line `summary` the consumer
+ * turns into a marker header (`[extracted text of <path> — <summary>]`) so the
+ * reader knows it is looking at extracted text, not the file's bytes.
+ *
+ * The `summary` (not a full marker) is what extractors return and what the
+ * cache stores, because the cache is keyed by CONTENT (git blob sha): the same
+ * document at two workspace paths shares one cache entry, and baking a path
+ * into the cached value would surface the wrong path on the second read.
+ */
+/**
+ * Build the honest ONE-LINE header consumers prepend to extracted text.
+ *
+ * One line is a promise the rest of the read path keeps: grep counts on the
+ * marker occupying exactly one, so read_file and grep agree on line numbers.
+ * A path may legally carry a CR or LF, which would forge extra lines and shift
+ * every number after it, so those are shown as escapes rather than obeyed.
+ */
+export function extractionMarker(path, summary) {
+    const oneLine = path.replace(/[\r\n]/g, (c) => (c === '\r' ? '\\r' : '\\n'));
+    return `[extracted text of ${oneLine} — ${summary}]`;
+}
+/** Lowercased extension of `path` including the dot, or '' when there is none. */
+export function fileExtension(path) {
+    const name = path.slice(path.lastIndexOf('/') + 1);
+    const dot = name.lastIndexOf('.');
+    return dot > 0 ? name.slice(dot).toLowerCase() : '';
+}
+//# sourceMappingURL=doc-extract.types.js.map

package/dist/modules/workspace/file-readers/doc-extract.types.js.map

@@ -0,0 +1 @@
+{"version":3,"file":"doc-extract.types.js","sourceRoot":"","sources":["../../../../src/modules/workspace/file-readers/doc-extract.types.ts"],"names":[],"mappings":"AAAA;;;;;;;;;;;;;;GAcG;AA8BH;;;;;;;GAOG;AACH,MAAM,UAAU,gBAAgB,CAAC,IAAY,EAAE,OAAe;IAC5D,MAAM,OAAO,GAAG,IAAI,CAAC,OAAO,CAAC,SAAS,EAAE,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC,CAAC,KAAK,IAAI,CAAC,CAAC,CAAC,KAAK,CAAC,CAAC,CAAC,KAAK,CAAC,CAAC,CAAC;IAC7E,OAAO,sBAAsB,OAAO,MAAM,OAAO,GAAG,CAAC;AACvD,CAAC;AAED,kFAAkF;AAClF,MAAM,UAAU,aAAa,CAAC,IAAY;IACxC,MAAM,IAAI,GAAG,IAAI,CAAC,KAAK,CAAC,IAAI,CAAC,WAAW,CAAC,GAAG,CAAC,GAAG,CAAC,CAAC,CAAC;IACnD,MAAM,GAAG,GAAG,IAAI,CAAC,WAAW,CAAC,GAAG,CAAC,CAAC;IAClC,OAAO,GAAG,GAAG,CAAC,CAAC,CAAC,CAAC,IAAI,CAAC,KAAK,CAAC,GAAG,CAAC,CAAC,WAAW,EAAE,CAAC,CAAC,CAAC,EAAE,CAAC;AACtD,CAAC"}

package/dist/modules/workspace/file-readers/document-reader.d.ts

@@ -0,0 +1,32 @@
+import type { DocExtractService } from './doc-extract.service.js';
+import type { ExtractFn } from './doc-extract.types.js';
+import { type FileReader, type ReadResult } from './file-reader.js';
+/**
+ * FileReader over one document format: a thin wrapper pairing the format's
+ * PURE extract function (extract-docx.ts and friends) with the shared
+ * content-hash extraction cache (`DocExtractService`). Reads return the
+ * extraction under its honest `[extracted text of …]` marker; a parse failure
+ * becomes a refusal message, never a 500.
+ *
+ * `grep` semantics: `greppableText` serves only the CACHED extraction (a hit
+ * costs one small JSON read), returning null for a cold document — whether to
+ * spend grep's per-walk extraction budget on a cold one is the walk's call,
+ * which then extracts through `read` (see grepWalk in workspace.tools.ts).
+ * grep also branches on `instanceof DocumentReader` for exactly that decision.
+ */
+export declare class DocumentReader implements FileReader {
+    private readonly extract;
+    private readonly service;
+    readonly extensions: readonly string[];
+    /**
+     * `read_file` returns an EXTRACTION for these types, so text written back
+     * could not round-trip — the write tools refuse (documents are replaced by
+     * uploading a new version).
+     */
+    readonly textEditable = false;
+    constructor(extension: string, extract: ExtractFn, service: DocExtractService);
+    read(bytes: Buffer, path: string): Promise<ReadResult>;
+    /** The cached extraction (marker line included, so grep's line numbers match read_file's), or null when cold. */
+    greppableText(bytes: Buffer, path: string): Promise<string | null>;
+}
+//# sourceMappingURL=document-reader.d.ts.map

package/dist/modules/workspace/file-readers/document-reader.d.ts.map

@@ -0,0 +1 @@
+{"version":3,"file":"document-reader.d.ts","sourceRoot":"","sources":["../../../../src/modules/workspace/file-readers/document-reader.ts"],"names":[],"mappings":"AAAA,OAAO,KAAK,EAAE,iBAAiB,EAAE,MAAM,0BAA0B,CAAC;AAClE,OAAO,KAAK,EAAE,SAAS,EAAE,MAAM,wBAAwB,CAAC;AACxD,OAAO,EAAwB,KAAK,UAAU,EAAE,KAAK,UAAU,EAAE,MAAM,kBAAkB,CAAC;AAE1F;;;;;;;;;;;;GAYG;AACH,qBAAa,cAAe,YAAW,UAAU;IAW7C,OAAO,CAAC,QAAQ,CAAC,OAAO;IACxB,OAAO,CAAC,QAAQ,CAAC,OAAO;IAX1B,QAAQ,CAAC,UAAU,EAAE,SAAS,MAAM,EAAE,CAAC;IACvC;;;;OAIG;IACH,QAAQ,CAAC,YAAY,SAAS;gBAG5B,SAAS,EAAE,MAAM,EACA,OAAO,EAAE,SAAS,EAClB,OAAO,EAAE,iBAAiB;IAKvC,IAAI,CAAC,KAAK,EAAE,MAAM,EAAE,IAAI,EAAE,MAAM,GAAG,OAAO,CAAC,UAAU,CAAC;IAwB5D,iHAAiH;IAC3G,aAAa,CAAC,KAAK,EAAE,MAAM,EAAE,IAAI,EAAE,MAAM,GAAG,OAAO,CAAC,MAAM,GAAG,IAAI,CAAC;CAIzE"}

package/dist/modules/workspace/file-readers/document-reader.js

@@ -0,0 +1,59 @@
+import { displayPath, oneLine } from './file-reader.js';
+/**
+ * FileReader over one document format: a thin wrapper pairing the format's
+ * PURE extract function (extract-docx.ts and friends) with the shared
+ * content-hash extraction cache (`DocExtractService`). Reads return the
+ * extraction under its honest `[extracted text of …]` marker; a parse failure
+ * becomes a refusal message, never a 500.
+ *
+ * `grep` semantics: `greppableText` serves only the CACHED extraction (a hit
+ * costs one small JSON read), returning null for a cold document — whether to
+ * spend grep's per-walk extraction budget on a cold one is the walk's call,
+ * which then extracts through `read` (see grepWalk in workspace.tools.ts).
+ * grep also branches on `instanceof DocumentReader` for exactly that decision.
+ */
+export class DocumentReader {
+    extract;
+    service;
+    extensions;
+    /**
+     * `read_file` returns an EXTRACTION for these types, so text written back
+     * could not round-trip — the write tools refuse (documents are replaced by
+     * uploading a new version).
+     */
+    textEditable = false;
+    constructor(extension, extract, service) {
+        this.extract = extract;
+        this.service = service;
+        this.extensions = [extension];
+    }
+    async read(bytes, path) {
+        // An extractor is contracted to ANSWER for bad content rather than throw,
+        // but it wraps third-party parsers, and one of those throwing past its own
+        // guard is a corrupt file — not a server fault. read_file says so instead
+        // of failing the whole tool call with a 500.
+        let res;
+        try {
+            res = await this.service.extract(path, bytes, this.extract);
+        }
+        catch (err) {
+            const reason = err instanceof Error ? err.message : String(err);
+            return {
+                kind: 'refusal',
+                message: `[${displayPath(path)} could not be extracted (${oneLine(reason)}) — the file may be corrupt or mislabeled. To fix it, replace the document by uploading a new version.]`,
+            };
+        }
+        return res.ok
+            ? { kind: 'text', text: `${res.marker}\n${res.text}` }
+            : {
+                kind: 'refusal',
+                message: `[${displayPath(path)} ${oneLine(res.message)} — the file may be corrupt or mislabeled. To fix it, replace the document by uploading a new version.]`,
+            };
+    }
+    /** The cached extraction (marker line included, so grep's line numbers match read_file's), or null when cold. */
+    async greppableText(bytes, path) {
+        const hit = await this.service.getCached(path, bytes);
+        return hit ? `${hit.marker}\n${hit.text}` : null;
+    }
+}
+//# sourceMappingURL=document-reader.js.map

package/dist/modules/workspace/file-readers/document-reader.js.map

@@ -0,0 +1 @@
+{"version":3,"file":"document-reader.js","sourceRoot":"","sources":["../../../../src/modules/workspace/file-readers/document-reader.ts"],"names":[],"mappings":"AAEA,OAAO,EAAE,WAAW,EAAE,OAAO,EAAoC,MAAM,kBAAkB,CAAC;AAE1F;;;;;;;;;;;;GAYG;AACH,MAAM,OAAO,cAAc;IAWN;IACA;IAXV,UAAU,CAAoB;IACvC;;;;OAIG;IACM,YAAY,GAAG,KAAK,CAAC;IAE9B,YACE,SAAiB,EACA,OAAkB,EAClB,OAA0B;QAD1B,YAAO,GAAP,OAAO,CAAW;QAClB,YAAO,GAAP,OAAO,CAAmB;QAE3C,IAAI,CAAC,UAAU,GAAG,CAAC,SAAS,CAAC,CAAC;IAChC,CAAC;IAED,KAAK,CAAC,IAAI,CAAC,KAAa,EAAE,IAAY;QACpC,0EAA0E;QAC1E,2EAA2E;QAC3E,0EAA0E;QAC1E,6CAA6C;QAC7C,IAAI,GAAqD,CAAC;QAC1D,IAAI,CAAC;YACH,GAAG,GAAG,MAAM,IAAI,CAAC,OAAO,CAAC,OAAO,CAAC,IAAI,EAAE,KAAK,EAAE,IAAI,CAAC,OAAO,CAAC,CAAC;QAC9D,CAAC;QAAC,OAAO,GAAG,EAAE,CAAC;YACb,MAAM,MAAM,GAAG,GAAG,YAAY,KAAK,CAAC,CAAC,CAAC,GAAG,CAAC,OAAO,CAAC,CAAC,CAAC,MAAM,CAAC,GAAG,CAAC,CAAC;YAChE,OAAO;gBACL,IAAI,EAAE,SAAS;gBACf,OAAO,EAAE,IAAI,WAAW,CAAC,IAAI,CAAC,4BAA4B,OAAO,CAAC,MAAM,CAAC,yGAAyG;aACnL,CAAC;QACJ,CAAC;QAED,OAAO,GAAG,CAAC,EAAE;YACX,CAAC,CAAC,EAAE,IAAI,EAAE,MAAM,EAAE,IAAI,EAAE,GAAG,GAAG,CAAC,MAAM,KAAK,GAAG,CAAC,IAAI,EAAE,EAAE;YACtD,CAAC,CAAC;gBACE,IAAI,EAAE,SAAS;gBACf,OAAO,EAAE,IAAI,WAAW,CAAC,IAAI,CAAC,IAAI,OAAO,CAAC,GAAG,CAAC,OAAO,CAAC,wGAAwG;aAC/J,CAAC;IACR,CAAC;IAED,iHAAiH;IACjH,KAAK,CAAC,aAAa,CAAC,KAAa,EAAE,IAAY;QAC7C,MAAM,GAAG,GAAG,MAAM,IAAI,CAAC,OAAO,CAAC,SAAS,CAAC,IAAI,EAAE,KAAK,CAAC,CAAC;QACtD,OAAO,GAAG,CAAC,CAAC,CAAC,GAAG,GAAG,CAAC,MAAM,KAAK,GAAG,CAAC,IAAI,EAAE,CAAC,CAAC,CAAC,IAAI,CAAC;IACnD,CAAC;CACF"}

package/dist/modules/workspace/file-readers/email-reader.d.ts

@@ -0,0 +1,15 @@
+import { DocumentReader } from './document-reader.js';
+/**
+ * FileReader over one email format (`.eml` / `.msg`): a `DocumentReader` in
+ * every mechanical respect — cache-wrapped extraction, greppable when cached,
+ * corrupt files answered with the typed could-not-be-parsed refusal — with
+ * only the WRITE-refusal copy specialized. The generic document refusal talks
+ * about round-tripping an office document; an email deserves the honest
+ * version of the same "no": the file is a snapshot of a message, and editing
+ * a snapshot's text is not a thing.
+ */
+export declare class EmailReader extends DocumentReader {
+    /** The write-refusal for the agent text-editing tools (see `assertNotDocumentEdit`). */
+    editRefusal(path: string): string;
+}
+//# sourceMappingURL=email-reader.d.ts.map

package/dist/modules/workspace/file-readers/email-reader.d.ts.map

@@ -0,0 +1 @@
+{"version":3,"file":"email-reader.d.ts","sourceRoot":"","sources":["../../../../src/modules/workspace/file-readers/email-reader.ts"],"names":[],"mappings":"AAAA,OAAO,EAAE,cAAc,EAAE,MAAM,sBAAsB,CAAC;AAEtD;;;;;;;;GAQG;AACH,qBAAa,WAAY,SAAQ,cAAc;IAC7C,wFAAwF;IACxF,WAAW,CAAC,IAAI,EAAE,MAAM,GAAG,MAAM;CAOlC"}

package/dist/modules/workspace/file-readers/email-reader.js

@@ -0,0 +1,19 @@
+import { DocumentReader } from './document-reader.js';
+/**
+ * FileReader over one email format (`.eml` / `.msg`): a `DocumentReader` in
+ * every mechanical respect — cache-wrapped extraction, greppable when cached,
+ * corrupt files answered with the typed could-not-be-parsed refusal — with
+ * only the WRITE-refusal copy specialized. The generic document refusal talks
+ * about round-tripping an office document; an email deserves the honest
+ * version of the same "no": the file is a snapshot of a message, and editing
+ * a snapshot's text is not a thing.
+ */
+export class EmailReader extends DocumentReader {
+    /** The write-refusal for the agent text-editing tools (see `assertNotDocumentEdit`). */
+    editRefusal(path) {
+        return (`"${path}" is an email file — a snapshot of a message. read_file returns EXTRACTED text for it, ` +
+            'and a snapshot cannot be text-edited; to change what is stored, replace the file by uploading ' +
+            'a new version.');
+    }
+}
+//# sourceMappingURL=email-reader.js.map

package/dist/modules/workspace/file-readers/email-reader.js.map

@@ -0,0 +1 @@
+{"version":3,"file":"email-reader.js","sourceRoot":"","sources":["../../../../src/modules/workspace/file-readers/email-reader.ts"],"names":[],"mappings":"AAAA,OAAO,EAAE,cAAc,EAAE,MAAM,sBAAsB,CAAC;AAEtD;;;;;;;;GAQG;AACH,MAAM,OAAO,WAAY,SAAQ,cAAc;IAC7C,wFAAwF;IACxF,WAAW,CAAC,IAAY;QACtB,OAAO,CACL,IAAI,IAAI,yFAAyF;YACjG,gGAAgG;YAChG,gBAAgB,CACjB,CAAC;IACJ,CAAC;CACF"}

package/dist/modules/workspace/file-readers/email-text.d.ts

@@ -0,0 +1,51 @@
+/**
+ * Shared shaping for the email extractors (`extract-eml.ts` / `extract-msg.ts`).
+ *
+ * Both formats extract to the SAME text shape, so an agent greps a mailbox
+ * without caring which client saved the file:
+ *
+ *   [from] Ada Lovelace <ada@example.com>
+ *   [to] Bob <bob@example.com>, carol@example.com
+ *   [subject] Quarterly numbers
+ *   [date] 2026-01-05T10:00:00.000Z
+ *
+ *   the body…
+ *
+ *   [attachments]
+ *   report.pdf (application/pdf, 48211 bytes)
+ *
+ * Header lines are omitted when the message lacks the field (never printed
+ * empty). The body prefers the plain-text part; an HTML-only body is stripped
+ * to text (block tags become newlines so paragraphs survive) and the marker
+ * summary says so. Attachments are LISTED by name only — v1 does not extract
+ * inside them, and the summary says that too.
+ */
+import type { ExtractedDoc } from './doc-extract.types.js';
+/** One listed attachment. Size/type are printed only when known. */
+export interface EmailAttachment {
+    name: string;
+    mimeType?: string;
+    sizeBytes?: number;
+}
+/** Where the body text came from — drives the honest summary + body notes. */
+export type EmailBodySource = 'text' | 'html' | 'rtf-only' | 'none';
+/** The format-independent email, as far as the extraction cares. */
+export interface EmailModel {
+    from?: string;
+    to?: string;
+    cc?: string;
+    bcc?: string;
+    subject?: string;
+    /** ISO timestamp when the date parsed, the raw header value otherwise. */
+    date?: string;
+    /** Body as plain text ('' when there is none or it is RTF-only). */
+    body: string;
+    bodySource: EmailBodySource;
+    attachments: EmailAttachment[];
+}
+/** The line a body that exists only as RTF gets INSTEAD of body text. */
+export declare const RTF_ONLY_BODY_LINE = "[body is RTF; no plain-text part]";
+export declare function htmlToEmailText(html: string): string;
+/** Render the model into the marker summary + extraction text (see module doc). */
+export declare function emailExtraction(model: EmailModel): ExtractedDoc;
+//# sourceMappingURL=email-text.d.ts.map

package/dist/modules/workspace/file-readers/email-text.d.ts.map

@@ -0,0 +1 @@
+{"version":3,"file":"email-text.d.ts","sourceRoot":"","sources":["../../../../src/modules/workspace/file-readers/email-text.ts"],"names":[],"mappings":"AAAA;;;;;;;;;;;;;;;;;;;;;GAqBG;AACH,OAAO,KAAK,EAAE,YAAY,EAAE,MAAM,wBAAwB,CAAC;AAI3D,oEAAoE;AACpE,MAAM,WAAW,eAAe;IAC9B,IAAI,EAAE,MAAM,CAAC;IACb,QAAQ,CAAC,EAAE,MAAM,CAAC;IAClB,SAAS,CAAC,EAAE,MAAM,CAAC;CACpB;AAED,8EAA8E;AAC9E,MAAM,MAAM,eAAe,GAAG,MAAM,GAAG,MAAM,GAAG,UAAU,GAAG,MAAM,CAAC;AAEpE,oEAAoE;AACpE,MAAM,WAAW,UAAU;IACzB,IAAI,CAAC,EAAE,MAAM,CAAC;IACd,EAAE,CAAC,EAAE,MAAM,CAAC;IACZ,EAAE,CAAC,EAAE,MAAM,CAAC;IACZ,GAAG,CAAC,EAAE,MAAM,CAAC;IACb,OAAO,CAAC,EAAE,MAAM,CAAC;IACjB,0EAA0E;IAC1E,IAAI,CAAC,EAAE,MAAM,CAAC;IACd,oEAAoE;IACpE,IAAI,EAAE,MAAM,CAAC;IACb,UAAU,EAAE,eAAe,CAAC;IAC5B,WAAW,EAAE,eAAe,EAAE,CAAC;CAChC;AAED,yEAAyE;AACzE,eAAO,MAAM,kBAAkB,sCAAsC,CAAC;AAsCtE,wBAAgB,eAAe,CAAC,IAAI,EAAE,MAAM,GAAG,MAAM,CA8CpD;AAsBD,mFAAmF;AACnF,wBAAgB,eAAe,CAAC,KAAK,EAAE,UAAU,GAAG,YAAY,CAiC/D"}

package/dist/modules/workspace/file-readers/email-text.js

@@ -0,0 +1,151 @@
+import { Parser } from 'htmlparser2';
+import { MAX_ELEMENT_DEPTH, TOO_DEEP } from './ooxml-text.js';
+/** The line a body that exists only as RTF gets INSTEAD of body text. */
+export const RTF_ONLY_BODY_LINE = '[body is RTF; no plain-text part]';
+/**
+ * Strip an HTML email body to plain text. Deliberately simple (the same
+ * stance as `ooxml-text.ts`): comments and `<style>`/`<script>`/`<head>`/
+ * `<title>` containers are dropped whole, `<br>` and block-level tag
+ * boundaries become newlines so paragraphs survive, every other tag is
+ * removed, and entities are decoded through the module's shared
+ * `decodeXmlEntities` (plus `&nbsp;`, which HTML has and XML does not). Runs
+ * of blank lines collapse to one.
+ *
+ * Implemented as a SINGLE-PASS linear scanner, not regexes: the earlier
+ * quote-aware tag regexes re-scanned the remaining body from every `<` when a
+ * quoted attribute never closed — malformed input with many `<` characters
+ * plus one unterminated quote pinned the server quadratically. The scanner is
+ * quote-aware the same way (a `>` INSIDE a quoted attribute value, as in
+ * `<a title="a > b">`, never ends the tag early) but amortizes the failures:
+ * a scan that reaches end-of-input marks every `<` it passed OUTSIDE quotes
+ * as known-literal (their scans would be identical tails), so no position is
+ * rescanned from more than the three possible quote states. An unterminated
+ * tag is literal text, not a tag — and so is a `<…>` span that names no
+ * element at all, which is how `1 < 2 > 0` survives into the body.
+ */
+/** Elements whose CONTENT is not body text and is dropped with the element. */
+const CONTAINER_TAGS = new Set(['script', 'style', 'head', 'title']);
+/** Elements whose boundaries end a line, so paragraphs survive as paragraphs. */
+const BLOCK_TAGS = new Set([
+    'p', 'div', 'section', 'article', 'header', 'footer', 'main',
+    'table', 'thead', 'tbody', 'tfoot', 'tr', 'td', 'th',
+    'li', 'ul', 'ol', 'dl', 'dt', 'dd', 'blockquote', 'pre', 'hr',
+    'h1', 'h2', 'h3', 'h4', 'h5', 'h6',
+]);
+// Depth bound shared with the OOXML/ODF extractors (`ooxml-text.ts`): mail
+// nests a few dozen levels even at its most table-happy; a crafted body can
+// nest as deep as it has bytes. What was read before the bound is kept.
+export function htmlToEmailText(html) {
+    let s = '';
+    let depth = 0;
+    let skipping = 0; // inside a container whose content is not body text
+    const parser = new Parser({
+        onopentag(name) {
+            depth++;
+            if (CONTAINER_TAGS.has(name))
+                skipping++;
+            else if (name === 'br' || BLOCK_TAGS.has(name))
+                s += '\n';
+            if (depth > MAX_ELEMENT_DEPTH)
+                throw TOO_DEEP;
+        },
+        ontext(text) {
+            if (skipping === 0)
+                s += text;
+        },
+        onclosetag(name) {
+            if (CONTAINER_TAGS.has(name)) {
+                if (skipping > 0)
+                    skipping--;
+            }
+            else if (BLOCK_TAGS.has(name))
+                s += '\n';
+            depth--;
+        },
+    }, 
+    // HTML mode, not XML: an email body is HTML, with its void elements, its
+    // implied closes, its raw-text `<script>`, and its named entities.
+    { decodeEntities: true });
+    try {
+        parser.write(html);
+        parser.end();
+    }
+    catch (err) {
+        if (err !== TOO_DEEP)
+            throw err;
+        parser.reset();
+    }
+    // `&nbsp;` decodes to U+00A0, which LOOKS like a space and is not one: an
+    // agent grepping the extraction for "Para one" would miss a line that reads
+    // exactly that. Extracted text is for reading and searching, so the
+    // non-breaking space becomes an ordinary one.
+    const out = [];
+    for (const raw of s.replace(/ /g, ' ').split('\n')) {
+        const line = raw.trim();
+        if (line !== '')
+            out.push(line);
+        else if (out.length > 0 && out[out.length - 1] !== '')
+            out.push('');
+    }
+    while (out.length > 0 && out[out.length - 1] === '')
+        out.pop();
+    return out.join('\n');
+}
+/** `name (mimeType, N bytes)` with the parenthesis dropped when nothing is known.
+ * CR/LF in the metadata are shown as escapes rather than obeyed, so each
+ * attachment stays on ONE line and grep line numbers hold. */
+function attachmentLine(a) {
+    // The extractors default a MISSING name, but an EMPTY (or whitespace-only)
+    // one reaches here as ''; without the same fallback the section printed a
+    // blank line — or one starting with the parenthesis — for that attachment.
+    const name = a.name.trim() === '' ? 'unnamed attachment' : a.name;
+    const details = [a.mimeType, a.sizeBytes !== undefined ? `${a.sizeBytes} bytes` : undefined]
+        .filter((d) => d !== undefined && d !== '')
+        .join(', ');
+    const line = details === '' ? name : `${name} (${details})`;
+    return line.replace(/[\r\n]/g, (c) => (c === '\r' ? '\\r' : '\\n'));
+}
+/** A header value with its line breaks made visible instead of obeyed. */
+function oneLine(value) {
+    return value.replace(/[\r\n]/g, (c) => (c === '\r' ? '\\r' : '\\n'));
+}
+/** Render the model into the marker summary + extraction text (see module doc). */
+export function emailExtraction(model) {
+    const header = [];
+    // Absent OR blank fields are omitted — a header marker is never printed empty.
+    const pushHeader = (label, value) => {
+        if (value === undefined || value.trim() === '')
+            return;
+        // A header marker is ONE line. A From or Subject carrying CR/LF would
+        // otherwise forge further lines into the extraction — including lines that
+        // read like other markers — so the breaks are shown as escapes.
+        header.push(`[${label}] ${oneLine(value)}`);
+    };
+    pushHeader('from', model.from);
+    pushHeader('to', model.to);
+    pushHeader('cc', model.cc);
+    pushHeader('bcc', model.bcc);
+    pushHeader('subject', model.subject);
+    pushHeader('date', model.date);
+    const sections = [];
+    if (header.length > 0)
+        sections.push(header.join('\n'));
+    if (model.bodySource === 'rtf-only')
+        sections.push(RTF_ONLY_BODY_LINE);
+    else if (model.body !== '')
+        sections.push(model.body);
+    if (model.attachments.length > 0) {
+        sections.push(`[attachments]\n${model.attachments.map(attachmentLine).join('\n')}`);
+    }
+    const n = model.attachments.length;
+    const parts = ['email message'];
+    if (n > 0)
+        parts.push(`${n} attachment${n === 1 ? '' : 's'} listed (names only; not extracted)`);
+    if (model.bodySource === 'html')
+        parts.push('HTML body rendered as plain text');
+    if (model.bodySource === 'rtf-only')
+        parts.push('body is RTF; no plain-text part');
+    parts.push('formatting and full headers omitted');
+    return { summary: parts.join('; '), text: sections.join('\n\n') };
+}
+//# sourceMappingURL=email-text.js.map

package/dist/modules/workspace/file-readers/email-text.js.map

@@ -0,0 +1 @@
+{"version":3,"file":"email-text.js","sourceRoot":"","sources":["../../../../src/modules/workspace/file-readers/email-text.ts"],"names":[],"mappings":"AAuBA,OAAO,EAAE,MAAM,EAAE,MAAM,aAAa,CAAC;AACrC,OAAO,EAAE,iBAAiB,EAAE,QAAQ,EAAE,MAAM,iBAAiB,CAAC;AA2B9D,yEAAyE;AACzE,MAAM,CAAC,MAAM,kBAAkB,GAAG,mCAAmC,CAAC;AAEtE;;;;;;;;;;;;;;;;;;;;GAoBG;AACH,+EAA+E;AAC/E,MAAM,cAAc,GAAG,IAAI,GAAG,CAAC,CAAC,QAAQ,EAAE,OAAO,EAAE,MAAM,EAAE,OAAO,CAAC,CAAC,CAAC;AAErE,iFAAiF;AACjF,MAAM,UAAU,GAAG,IAAI,GAAG,CAAC;IACzB,GAAG,EAAE,KAAK,EAAE,SAAS,EAAE,SAAS,EAAE,QAAQ,EAAE,QAAQ,EAAE,MAAM;IAC5D,OAAO,EAAE,OAAO,EAAE,OAAO,EAAE,OAAO,EAAE,IAAI,EAAE,IAAI,EAAE,IAAI;IACpD,IAAI,EAAE,IAAI,EAAE,IAAI,EAAE,IAAI,EAAE,IAAI,EAAE,IAAI,EAAE,YAAY,EAAE,KAAK,EAAE,IAAI;IAC7D,IAAI,EAAE,IAAI,EAAE,IAAI,EAAE,IAAI,EAAE,IAAI,EAAE,IAAI;CACnC,CAAC,CAAC;AAEH,2EAA2E;AAC3E,4EAA4E;AAC5E,wEAAwE;AAExE,MAAM,UAAU,eAAe,CAAC,IAAY;IAC1C,IAAI,CAAC,GAAG,EAAE,CAAC;IACX,IAAI,KAAK,GAAG,CAAC,CAAC;IACd,IAAI,QAAQ,GAAG,CAAC,CAAC,CAAC,oDAAoD;IACtE,MAAM,MAAM,GAAG,IAAI,MAAM,CACvB;QACE,SAAS,CAAC,IAAI;YACZ,KAAK,EAAE,CAAC;YACR,IAAI,cAAc,CAAC,GAAG,CAAC,IAAI,CAAC;gBAAE,QAAQ,EAAE,CAAC;iBACpC,IAAI,IAAI,KAAK,IAAI,IAAI,UAAU,CAAC,GAAG,CAAC,IAAI,CAAC;gBAAE,CAAC,IAAI,IAAI,CAAC;YAC1D,IAAI,KAAK,GAAG,iBAAiB;gBAAE,MAAM,QAAQ,CAAC;QAChD,CAAC;QACD,MAAM,CAAC,IAAI;YACT,IAAI,QAAQ,KAAK,CAAC;gBAAE,CAAC,IAAI,IAAI,CAAC;QAChC,CAAC;QACD,UAAU,CAAC,IAAI;YACb,IAAI,cAAc,CAAC,GAAG,CAAC,IAAI,CAAC,EAAE,CAAC;gBAC7B,IAAI,QAAQ,GAAG,CAAC;oBAAE,QAAQ,EAAE,CAAC;YAC/B,CAAC;iBAAM,IAAI,UAAU,CAAC,GAAG,CAAC,IAAI,CAAC;gBAAE,CAAC,IAAI,IAAI,CAAC;YAC3C,KAAK,EAAE,CAAC;QACV,CAAC;KACF;IACD,yEAAyE;IACzE,mEAAmE;IACnE,EAAE,cAAc,EAAE,IAAI,EAAE,CACzB,CAAC;IACF,IAAI,CAAC;QACH,MAAM,CAAC,KAAK,CAAC,IAAI,CAAC,CAAC;QACnB,MAAM,CAAC,GAAG,EAAE,CAAC;IACf,CAAC;IAAC,OAAO,GAAG,EAAE,CAAC;QACb,IAAI,GAAG,KAAK,QAAQ;YAAE,MAAM,GAAG,CAAC;QAChC,MAAM,CAAC,KAAK,EAAE,CAAC;IACjB,CAAC;IAED,0EAA0E;IAC1E,4EAA4E;IAC5E,oEAAoE;IACpE,8CAA8C;IAC9C,MAAM,GAAG,GAAa,EAAE,CAAC;IACzB,KAAK,MAAM,GAAG,IAAI,CAAC,CAAC,OAAO,CAAC,IAAI,EAAE,GAAG,CAAC,CAAC,KAAK,CAAC,IAAI,CAAC,EAAE,CAAC;QACnD,MAAM,IAAI,GAAG,GAAG,CAAC,IAAI,EAAE,CAAC;QACxB,IAAI,IAAI,KAAK,EAAE;YAAE,GAAG,CAAC,IAAI,CAAC,IAAI,CAAC,CAAC;aAC3B,IAAI,GAAG,CAAC,MAAM,GAAG,CAAC,IAAI,GAAG,CAAC,GAAG,CAAC,MAAM,GAAG,CAAC,CAAC,KAAK,EAAE;YAAE,GAAG,CAAC,IAAI,CAAC,EAAE,CAAC,CAAC;IACtE,CAAC;IACD,OAAO,GAAG,CAAC,MAAM,GAAG,CAAC,IAAI,GAAG,CAAC,GAAG,CAAC,MAAM,GAAG,CAAC,CAAC,KAAK,EAAE;QAAE,GAAG,CAAC,GAAG,EAAE,CAAC;IAC/D,OAAO,GAAG,CAAC,IAAI,CAAC,IAAI,CAAC,CAAC;AACxB,CAAC;AAED;;8DAE8D;AAC9D,SAAS,cAAc,CAAC,CAAkB;IACxC,2EAA2E;IAC3E,0EAA0E;IAC1E,2EAA2E;IAC3E,MAAM,IAAI,GAAG,CAAC,CAAC,IAAI,CAAC,IAAI,EAAE,KAAK,EAAE,CAAC,CAAC,CAAC,oBAAoB,CAAC,CAAC,CAAC,CAAC,CAAC,IAAI,CAAC;IAClE,MAAM,OAAO,GAAG,CAAC,CAAC,CAAC,QAAQ,EAAE,CAAC,CAAC,SAAS,KAAK,SAAS,CAAC,CAAC,CAAC,GAAG,CAAC,CAAC,SAAS,QAAQ,CAAC,CAAC,CAAC,SAAS,CAAC;SACzF,MAAM,CAAC,CAAC,CAAC,EAAe,EAAE,CAAC,CAAC,KAAK,SAAS,IAAI,CAAC,KAAK,EAAE,CAAC;SACvD,IAAI,CAAC,IAAI,CAAC,CAAC;IACd,MAAM,IAAI,GAAG,OAAO,KAAK,EAAE,CAAC,CAAC,CAAC,IAAI,CAAC,CAAC,CAAC,GAAG,IAAI,KAAK,OAAO,GAAG,CAAC;IAC5D,OAAO,IAAI,CAAC,OAAO,CAAC,SAAS,EAAE,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC,CAAC,KAAK,IAAI,CAAC,CAAC,CAAC,KAAK,CAAC,CAAC,CAAC,KAAK,CAAC,CAAC,CAAC;AACtE,CAAC;AAED,0EAA0E;AAC1E,SAAS,OAAO,CAAC,KAAa;IAC5B,OAAO,KAAK,CAAC,OAAO,CAAC,SAAS,EAAE,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC,CAAC,KAAK,IAAI,CAAC,CAAC,CAAC,KAAK,CAAC,CAAC,CAAC,KAAK,CAAC,CAAC,CAAC;AACvE,CAAC;AAED,mFAAmF;AACnF,MAAM,UAAU,eAAe,CAAC,KAAiB;IAC/C,MAAM,MAAM,GAAa,EAAE,CAAC;IAC5B,+EAA+E;IAC/E,MAAM,UAAU,GAAG,CAAC,KAAa,EAAE,KAAyB,EAAQ,EAAE;QACpE,IAAI,KAAK,KAAK,SAAS,IAAI,KAAK,CAAC,IAAI,EAAE,KAAK,EAAE;YAAE,OAAO;QACvD,sEAAsE;QACtE,2EAA2E;QAC3E,gEAAgE;QAChE,MAAM,CAAC,IAAI,CAAC,IAAI,KAAK,KAAK,OAAO,CAAC,KAAK,CAAC,EAAE,CAAC,CAAC;IAC9C,CAAC,CAAC;IACF,UAAU,CAAC,MAAM,EAAE,KAAK,CAAC,IAAI,CAAC,CAAC;IAC/B,UAAU,CAAC,IAAI,EAAE,KAAK,CAAC,EAAE,CAAC,CAAC;IAC3B,UAAU,CAAC,IAAI,EAAE,KAAK,CAAC,EAAE,CAAC,CAAC;IAC3B,UAAU,CAAC,KAAK,EAAE,KAAK,CAAC,GAAG,CAAC,CAAC;IAC7B,UAAU,CAAC,SAAS,EAAE,KAAK,CAAC,OAAO,CAAC,CAAC;IACrC,UAAU,CAAC,MAAM,EAAE,KAAK,CAAC,IAAI,CAAC,CAAC;IAE/B,MAAM,QAAQ,GAAa,EAAE,CAAC;IAC9B,IAAI,MAAM,CAAC,MAAM,GAAG,CAAC;QAAE,QAAQ,CAAC,IAAI,CAAC,MAAM,CAAC,IAAI,CAAC,IAAI,CAAC,CAAC,CAAC;IACxD,IAAI,KAAK,CAAC,UAAU,KAAK,UAAU;QAAE,QAAQ,CAAC,IAAI,CAAC,kBAAkB,CAAC,CAAC;SAClE,IAAI,KAAK,CAAC,IAAI,KAAK,EAAE;QAAE,QAAQ,CAAC,IAAI,CAAC,KAAK,CAAC,IAAI,CAAC,CAAC;IACtD,IAAI,KAAK,CAAC,WAAW,CAAC,MAAM,GAAG,CAAC,EAAE,CAAC;QACjC,QAAQ,CAAC,IAAI,CAAC,kBAAkB,KAAK,CAAC,WAAW,CAAC,GAAG,CAAC,cAAc,CAAC,CAAC,IAAI,CAAC,IAAI,CAAC,EAAE,CAAC,CAAC;IACtF,CAAC;IAED,MAAM,CAAC,GAAG,KAAK,CAAC,WAAW,CAAC,MAAM,CAAC;IACnC,MAAM,KAAK,GAAG,CAAC,eAAe,CAAC,CAAC;IAChC,IAAI,CAAC,GAAG,CAAC;QAAE,KAAK,CAAC,IAAI,CAAC,GAAG,CAAC,cAAc,CAAC,KAAK,CAAC,CAAC,CAAC,CAAC,EAAE,CAAC,CAAC,CAAC,GAAG,qCAAqC,CAAC,CAAC;IACjG,IAAI,KAAK,CAAC,UAAU,KAAK,MAAM;QAAE,KAAK,CAAC,IAAI,CAAC,kCAAkC,CAAC,CAAC;IAChF,IAAI,KAAK,CAAC,UAAU,KAAK,UAAU;QAAE,KAAK,CAAC,IAAI,CAAC,iCAAiC,CAAC,CAAC;IACnF,KAAK,CAAC,IAAI,CAAC,qCAAqC,CAAC,CAAC;IAElD,OAAO,EAAE,OAAO,EAAE,KAAK,CAAC,IAAI,CAAC,IAAI,CAAC,EAAE,IAAI,EAAE,QAAQ,CAAC,IAAI,CAAC,MAAM,CAAC,EAAE,CAAC;AACpE,CAAC"}

package/dist/modules/workspace/file-readers/extract-docx.d.ts

@@ -0,0 +1,13 @@
+import type { ExtractResult } from './doc-extract.types.js';
+/**
+ * Extract the BODY text of a `.docx` (Word) document.
+ *
+ * A docx is a zip whose main part is `word/document.xml`. v1 extracts the body
+ * only — headers/footers are skipped, and the marker summary says so.
+ *
+ *  - Paragraphs become lines. `<w:t>` runs are concatenated with NO separator
+ *    (Word splits runs mid-word on formatting boundaries).
+ *  - Tables become lines with cell text tab-separated, one line per row.
+ */
+export declare function extractDocx(bytes: Buffer): ExtractResult;
+//# sourceMappingURL=extract-docx.d.ts.map

package/dist/modules/workspace/file-readers/extract-docx.d.ts.map

@@ -0,0 +1 @@
+{"version":3,"file":"extract-docx.d.ts","sourceRoot":"","sources":["../../../../src/modules/workspace/file-readers/extract-docx.ts"],"names":[],"mappings":"AACA,OAAO,KAAK,EAAE,aAAa,EAAE,MAAM,wBAAwB,CAAC;AAG5D;;;;;;;;;GASG;AACH,wBAAgB,WAAW,CAAC,KAAK,EAAE,MAAM,GAAG,aAAa,CAmDxD"}