@yadsh/dsh-qa-surface 0.4.0 → 0.6.0

package/lib/documents/providers/docling.js

@@ -0,0 +1,247 @@
+/**
+ * Docling provider — the default DOCX/PDF → Markdown extractor (§6.1, §15).
+ *
+ * Docling runs as a separate HTTP service (`docling-serve`) rather than inside
+ * the host process: it is the heaviest dependency of the pipeline, it is where
+ * OCR happens, and keeping it across a network boundary is what lets the QA
+ * image stay small (§21, §39).
+ *
+ * Two behaviours are deliberate:
+ *
+ * - images come back inline as base64 data URIs (the only representation that
+ *   survives an HTTP boundary without a second fetch), and the extractor writes
+ *   them out as real files under `assets/`, rewriting references to relative
+ *   paths (§17.2);
+ * - nothing about the response shape is assumed beyond "there is Markdown
+ *   somewhere in it": a deployment upgrades docling-serve independently of
+ *   this plugin, so both the wrapped (`document.md_content`) and flat
+ *   (`md_content`) spellings are read.
+ */
+import { mkdir } from "node:fs/promises";
+import path from "node:path";
+import { DocumentError, sanitizeBackendOutput } from "../errors.js";
+import { backendInfo } from "./shared.js";
+const HEALTH_TIMEOUT_MS = 5_000;
+const DATA_URI_PATTERN = /!\[([^\]]*)\]\(data:image\/(png|jpe?g|gif|webp|bmp|tiff);base64,([A-Za-z0-9+/=\s]+)\)/gu;
+const EXTENSION_BY_MIME = {
+    png: "png",
+    jpg: "jpeg",
+    jpeg: "jpeg",
+    gif: "gif",
+    webp: "webp",
+    bmp: "bmp",
+    tiff: "tiff",
+};
+function joinUrl(baseUrl, route) {
+    const trimmed = baseUrl.endsWith("/") ? baseUrl.slice(0, -1) : baseUrl;
+    return `${trimmed}${route}`;
+}
+export class DoclingExtractor {
+    options;
+    name = "docling";
+    constructor(options) {
+        this.options = options;
+    }
+    supports(_format) {
+        return true;
+    }
+    get fetchImpl() {
+        return this.options.fetchImpl ?? fetch;
+    }
+    async health() {
+        for (const route of ["/health", "/v1/health"]) {
+            try {
+                const response = await this.fetchImpl(joinUrl(this.options.baseUrl, route), {
+                    method: "GET",
+                    signal: AbortSignal.timeout(HEALTH_TIMEOUT_MS),
+                });
+                if (response.ok)
+                    return "ok";
+            }
+            catch {
+                continue;
+            }
+        }
+        return "unavailable";
+    }
+    async extract(input) {
+        const warnings = [];
+        const body = new FormData();
+        const file = await import("node:fs/promises").then((fs) => fs.readFile(input.inputPath));
+        body.append("files", new Blob([new Uint8Array(file)], { type: "application/octet-stream" }), path.basename(input.inputPath));
+        body.append("to_formats", "md");
+        body.append("image_export_mode", input.extractImages ? "embedded" : "placeholder");
+        body.append("abort_on_error", "false");
+        if (input.extractTables)
+            body.append("table_mode", "accurate");
+        if (input.ocr === "force") {
+            body.append("do_ocr", "true");
+            body.append("force_ocr", "true");
+            warnings.push({
+                code: "OCR_USED",
+                message: "Document was processed using OCR.",
+                backend: this.name,
+            });
+        }
+        else if (input.ocr === "off") {
+            body.append("do_ocr", "false");
+        }
+        else {
+            // Let Docling decide per page; the service turns OCR on for pages that
+            // carry no text layer.
+            body.append("do_ocr", "true");
+            body.append("force_ocr", "false");
+        }
+        for (const language of input.ocrLanguages ?? []) {
+            const clean = language.trim();
+            if (clean !== "")
+                body.append("ocr_lang", clean);
+        }
+        let response;
+        try {
+            response = await this.fetchImpl(joinUrl(this.options.baseUrl, "/v1/convert/file"), {
+                method: "POST",
+                body,
+                signal: input.signal ?? AbortSignal.timeout(this.options.timeoutMs),
+            });
+        }
+        catch (error) {
+            const aborted = input.signal?.aborted === true;
+            throw new DocumentError(aborted ? "BACKEND_TIMEOUT" : "BACKEND_UNAVAILABLE", aborted
+                ? "the document extractor was cancelled"
+                : `the document extractor is unreachable: ${sanitizeBackendOutput(error instanceof Error ? error.message : String(error))}`, { backend: this.name, cause: error });
+        }
+        if (!response.ok) {
+            const detail = sanitizeBackendOutput(await response.text().catch(() => ""));
+            throw new DocumentError(response.status === 503 ? "BACKEND_UNAVAILABLE" : "EXTRACTION_FAILED", `the document extractor answered ${response.status}${detail === "" ? "" : `: ${detail}`}`, { backend: this.name, details: { status: response.status } });
+        }
+        const raw = await response.text();
+        if (raw.length > this.options.maxResponseBytes) {
+            throw new DocumentError("DOCUMENT_TOO_LARGE", "the extracted document exceeds the configured response budget", { backend: this.name, details: { bytes: raw.length } });
+        }
+        let payload;
+        try {
+            payload = JSON.parse(raw);
+        }
+        catch (error) {
+            throw new DocumentError("EXTRACTION_FAILED", "the document extractor returned a response that is not JSON", { backend: this.name, cause: error });
+        }
+        const { markdown, pages, serviceVersion } = readDoclingPayload(payload, this.name);
+        const { text, assets } = await this.materializeAssets(markdown, input.assetsDir, input.extractImages, warnings);
+        if (input.preservePageMarkers && !/\f/u.test(text)) {
+            warnings.push({
+                code: "METADATA_PARTIALLY_EXTRACTED",
+                message: "the extractor reported no page boundaries for this document",
+                backend: this.name,
+            });
+        }
+        return {
+            markdown: text,
+            assets,
+            ...(pages === undefined ? {} : { pages }),
+            backend: backendInfo(this.name, serviceVersion),
+            warnings,
+        };
+    }
+    /**
+     * Write inline images out as files and rewrite the Markdown to reference
+     * them relatively. Images beyond the configured cap are dropped with a
+     * warning rather than silently inflating the artifact (§25).
+     */
+    async materializeAssets(markdown, assetsDir, extractImages, warnings) {
+        const matches = [...markdown.matchAll(DATA_URI_PATTERN)];
+        if (matches.length === 0) {
+            if (extractImages && markdown.includes("data:image/")) {
+                warnings.push({
+                    code: "IMAGE_SKIPPED",
+                    message: "images reported by the extractor could not be decoded",
+                    backend: this.name,
+                });
+            }
+            return { text: markdown, assets: [] };
+        }
+        await mkdir(assetsDir, { recursive: true });
+        const assets = [];
+        const replacements = new Map();
+        let index = 0;
+        let skipped = 0;
+        const { writeFile } = await import("node:fs/promises");
+        for (const match of matches) {
+            const [full, alt = "", mime = "png", base64 = ""] = match;
+            if (index >= this.options.maxImages) {
+                skipped += 1;
+                replacements.set(full, alt);
+                continue;
+            }
+            const extension = EXTENSION_BY_MIME[mime.toLowerCase()] ?? "bin";
+            const fileName = `image-${String(index + 1).padStart(3, "0")}.${extension}`;
+            const payload = Buffer.from(base64.replace(/\s+/gu, ""), "base64");
+            if (payload.length === 0) {
+                skipped += 1;
+                replacements.set(full, alt);
+                continue;
+            }
+            await writeFile(path.join(assetsDir, fileName), payload);
+            assets.push({
+                path: fileName,
+                mediaType: `image/${extension === "jpg" ? "jpeg" : extension}`,
+            });
+            replacements.set(full, `![${alt}](assets/${fileName})`);
+            index += 1;
+        }
+        let text = markdown;
+        for (const [from, to] of replacements) {
+            text = text.split(from).join(to);
+        }
+        if (skipped > 0) {
+            warnings.push({
+                code: "IMAGE_SKIPPED",
+                message: `${skipped} image(s) were left out: the configured image budget is ${this.options.maxImages}`,
+                backend: this.name,
+                details: { skipped },
+            });
+        }
+        if (!extractImages && assets.length > 0) {
+            warnings.push({
+                code: "IMAGE_SKIPPED",
+                message: "extracted images were written anyway; extraction asked for none",
+                backend: this.name,
+            });
+        }
+        return { text, assets };
+    }
+}
+/** Read the Markdown out of a docling-serve response, in either spelling. */
+export function readDoclingPayload(payload, backend) {
+    if (payload === null || typeof payload !== "object") {
+        throw new DocumentError("EXTRACTION_FAILED", "the extractor returned no document", {
+            backend,
+        });
+    }
+    const root = payload;
+    const status = root["status"];
+    if (typeof status === "string" && status.toLowerCase() !== "success") {
+        const errors = Array.isArray(root["errors"])
+            ? root["errors"]
+                .map((entry) => sanitizeBackendOutput(JSON.stringify(entry), 300))
+                .join("; ")
+            : "";
+        throw new DocumentError("EXTRACTION_FAILED", `the extractor reported "${status}"${errors === "" ? "" : `: ${errors}`}`, { backend, details: { status } });
+    }
+    const document = root["document"] !== null && typeof root["document"] === "object"
+        ? root["document"]
+        : root;
+    const markdown = document["md_content"] ?? document["markdown"];
+    if (typeof markdown !== "string") {
+        throw new DocumentError("EXTRACTION_FAILED", "the extractor returned no Markdown content", { backend });
+    }
+    const pagesSource = document["page_count"] ?? document["pages"] ?? root["page_count"];
+    const pages = typeof pagesSource === "number" ? pagesSource : undefined;
+    const version = root["version"] ?? document["version"];
+    return {
+        markdown,
+        ...(pages === undefined ? {} : { pages }),
+        ...(typeof version === "string" ? { serviceVersion: version } : {}),
+    };
+}
+//# sourceMappingURL=docling.js.map

package/lib/documents/providers/docling.js.map

@@ -0,0 +1 @@
+{"version":3,"file":"docling.js","sourceRoot":"","sources":["../../../src/documents/providers/docling.ts"],"names":[],"mappings":"AAAA;;;;;;;;;;;;;;;;;;GAkBG;AAEH,OAAO,EAAE,KAAK,EAAE,MAAM,kBAAkB,CAAC;AACzC,OAAO,IAAI,MAAM,WAAW,CAAC;AAE7B,OAAO,EAAE,aAAa,EAAE,qBAAqB,EAAE,MAAM,cAAc,CAAC;AASpE,OAAO,EAAE,WAAW,EAAE,MAAM,aAAa,CAAC;AAa1C,MAAM,iBAAiB,GAAG,KAAK,CAAC;AAChC,MAAM,gBAAgB,GACpB,yFAAyF,CAAC;AAE5F,MAAM,iBAAiB,GAAqC;IAC1D,GAAG,EAAE,KAAK;IACV,GAAG,EAAE,MAAM;IACX,IAAI,EAAE,MAAM;IACZ,GAAG,EAAE,KAAK;IACV,IAAI,EAAE,MAAM;IACZ,GAAG,EAAE,KAAK;IACV,IAAI,EAAE,MAAM;CACb,CAAC;AAEF,SAAS,OAAO,CAAC,OAAe,EAAE,KAAa;IAC7C,MAAM,OAAO,GAAG,OAAO,CAAC,QAAQ,CAAC,GAAG,CAAC,CAAC,CAAC,CAAC,OAAO,CAAC,KAAK,CAAC,CAAC,EAAE,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC,OAAO,CAAC;IACvE,OAAO,GAAG,OAAO,GAAG,KAAK,EAAE,CAAC;AAC9B,CAAC;AAED,MAAM,OAAO,gBAAgB;IAGE,OAAO;IAF3B,IAAI,GAAG,SAAS,CAAC;IAE1B,YAA6B,OAA+B;uBAA/B,OAAO;IAA2B,CAAC;IAEhE,QAAQ,CAAC,OAAuB;QAC9B,OAAO,IAAI,CAAC;IACd,CAAC;IAED,IAAY,SAAS;QACnB,OAAO,IAAI,CAAC,OAAO,CAAC,SAAS,IAAI,KAAK,CAAC;IACzC,CAAC;IAED,KAAK,CAAC,MAAM;QACV,KAAK,MAAM,KAAK,IAAI,CAAC,SAAS,EAAE,YAAY,CAAC,EAAE,CAAC;YAC9C,IAAI,CAAC;gBACH,MAAM,QAAQ,GAAG,MAAM,IAAI,CAAC,SAAS,CACnC,OAAO,CAAC,IAAI,CAAC,OAAO,CAAC,OAAO,EAAE,KAAK,CAAC,EACpC;oBACE,MAAM,EAAE,KAAK;oBACb,MAAM,EAAE,WAAW,CAAC,OAAO,CAAC,iBAAiB,CAAC;iBAC/C,CACF,CAAC;gBACF,IAAI,QAAQ,CAAC,EAAE;oBAAE,OAAO,IAAI,CAAC;YAC/B,CAAC;YAAC,MAAM,CAAC;gBACP,SAAS;YACX,CAAC;QACH,CAAC;QACD,OAAO,aAAa,CAAC;IACvB,CAAC;IAED,KAAK,CAAC,OAAO,CAAC,KAAmB;QAC/B,MAAM,QAAQ,GAAsB,EAAE,CAAC;QACvC,MAAM,IAAI,GAAG,IAAI,QAAQ,EAAE,CAAC;QAC5B,MAAM,IAAI,GAAG,MAAM,MAAM,CAAC,kBAAkB,CAAC,CAAC,IAAI,CAAC,CAAC,EAAE,EAAE,EAAE,CACxD,EAAE,CAAC,QAAQ,CAAC,KAAK,CAAC,SAAS,CAAC,CAC7B,CAAC;QACF,IAAI,CAAC,MAAM,CACT,OAAO,EACP,IAAI,IAAI,CAAC,CAAC,IAAI,UAAU,CAAC,IAAI,CAAC,CAAC,EAAE,EAAE,IAAI,EAAE,0BAA0B,EAAE,CAAC,EACtE,IAAI,CAAC,QAAQ,CAAC,KAAK,CAAC,SAAS,CAAC,CAC/B,CAAC;QACF,IAAI,CAAC,MAAM,CAAC,YAAY,EAAE,IAAI,CAAC,CAAC;QAChC,IAAI,CAAC,MAAM,CACT,mBAAmB,EACnB,KAAK,CAAC,aAAa,CAAC,CAAC,CAAC,UAAU,CAAC,CAAC,CAAC,aAAa,CACjD,CAAC;QACF,IAAI,CAAC,MAAM,CAAC,gBAAgB,EAAE,OAAO,CAAC,CAAC;QACvC,IAAI,KAAK,CAAC,aAAa;YAAE,IAAI,CAAC,MAAM,CAAC,YAAY,EAAE,UAAU,CAAC,CAAC;QAC/D,IAAI,KAAK,CAAC,GAAG,KAAK,OAAO,EAAE,CAAC;YAC1B,IAAI,CAAC,MAAM,CAAC,QAAQ,EAAE,MAAM,CAAC,CAAC;YAC9B,IAAI,CAAC,MAAM,CAAC,WAAW,EAAE,MAAM,CAAC,CAAC;YACjC,QAAQ,CAAC,IAAI,CAAC;gBACZ,IAAI,EAAE,UAAU;gBAChB,OAAO,EAAE,mCAAmC;gBAC5C,OAAO,EAAE,IAAI,CAAC,IAAI;aACnB,CAAC,CAAC;QACL,CAAC;aAAM,IAAI,KAAK,CAAC,GAAG,KAAK,KAAK,EAAE,CAAC;YAC/B,IAAI,CAAC,MAAM,CAAC,QAAQ,EAAE,OAAO,CAAC,CAAC;QACjC,CAAC;aAAM,CAAC;YACN,uEAAuE;YACvE,uBAAuB;YACvB,IAAI,CAAC,MAAM,CAAC,QAAQ,EAAE,MAAM,CAAC,CAAC;YAC9B,IAAI,CAAC,MAAM,CAAC,WAAW,EAAE,OAAO,CAAC,CAAC;QACpC,CAAC;QACD,KAAK,MAAM,QAAQ,IAAI,KAAK,CAAC,YAAY,IAAI,EAAE,EAAE,CAAC;YAChD,MAAM,KAAK,GAAG,QAAQ,CAAC,IAAI,EAAE,CAAC;YAC9B,IAAI,KAAK,KAAK,EAAE;gBAAE,IAAI,CAAC,MAAM,CAAC,UAAU,EAAE,KAAK,CAAC,CAAC;QACnD,CAAC;QAED,IAAI,QAAkB,CAAC;QACvB,IAAI,CAAC;YACH,QAAQ,GAAG,MAAM,IAAI,CAAC,SAAS,CAC7B,OAAO,CAAC,IAAI,CAAC,OAAO,CAAC,OAAO,EAAE,kBAAkB,CAAC,EACjD;gBACE,MAAM,EAAE,MAAM;gBACd,IAAI;gBACJ,MAAM,EAAE,KAAK,CAAC,MAAM,IAAI,WAAW,CAAC,OAAO,CAAC,IAAI,CAAC,OAAO,CAAC,SAAS,CAAC;aACpE,CACF,CAAC;QACJ,CAAC;QAAC,OAAO,KAAK,EAAE,CAAC;YACf,MAAM,OAAO,GAAG,KAAK,CAAC,MAAM,EAAE,OAAO,KAAK,IAAI,CAAC;YAC/C,MAAM,IAAI,aAAa,CACrB,OAAO,CAAC,CAAC,CAAC,iBAAiB,CAAC,CAAC,CAAC,qBAAqB,EACnD,OAAO;gBACL,CAAC,CAAC,sCAAsC;gBACxC,CAAC,CAAC,0CAA0C,qBAAqB,CAC7D,KAAK,YAAY,KAAK,CAAC,CAAC,CAAC,KAAK,CAAC,OAAO,CAAC,CAAC,CAAC,MAAM,CAAC,KAAK,CAAC,CACvD,EAAE,EACP,EAAE,OAAO,EAAE,IAAI,CAAC,IAAI,EAAE,KAAK,EAAE,KAAK,EAAE,CACrC,CAAC;QACJ,CAAC;QAED,IAAI,CAAC,QAAQ,CAAC,EAAE,EAAE,CAAC;YACjB,MAAM,MAAM,GAAG,qBAAqB,CAClC,MAAM,QAAQ,CAAC,IAAI,EAAE,CAAC,KAAK,CAAC,GAAG,EAAE,CAAC,EAAE,CAAC,CACtC,CAAC;YACF,MAAM,IAAI,aAAa,CACrB,QAAQ,CAAC,MAAM,KAAK,GAAG,CAAC,CAAC,CAAC,qBAAqB,CAAC,CAAC,CAAC,mBAAmB,EACrE,mCAAmC,QAAQ,CAAC,MAAM,GAAG,MAAM,KAAK,EAAE,CAAC,CAAC,CAAC,EAAE,CAAC,CAAC,CAAC,KAAK,MAAM,EAAE,EAAE,EACzF,EAAE,OAAO,EAAE,IAAI,CAAC,IAAI,EAAE,OAAO,EAAE,EAAE,MAAM,EAAE,QAAQ,CAAC,MAAM,EAAE,EAAE,CAC7D,CAAC;QACJ,CAAC;QAED,MAAM,GAAG,GAAG,MAAM,QAAQ,CAAC,IAAI,EAAE,CAAC;QAClC,IAAI,GAAG,CAAC,MAAM,GAAG,IAAI,CAAC,OAAO,CAAC,gBAAgB,EAAE,CAAC;YAC/C,MAAM,IAAI,aAAa,CACrB,oBAAoB,EACpB,+DAA+D,EAC/D,EAAE,OAAO,EAAE,IAAI,CAAC,IAAI,EAAE,OAAO,EAAE,EAAE,KAAK,EAAE,GAAG,CAAC,MAAM,EAAE,EAAE,CACvD,CAAC;QACJ,CAAC;QAED,IAAI,OAAgB,CAAC;QACrB,IAAI,CAAC;YACH,OAAO,GAAG,IAAI,CAAC,KAAK,CAAC,GAAG,CAAC,CAAC;QAC5B,CAAC;QAAC,OAAO,KAAK,EAAE,CAAC;YACf,MAAM,IAAI,aAAa,CACrB,mBAAmB,EACnB,6DAA6D,EAC7D,EAAE,OAAO,EAAE,IAAI,CAAC,IAAI,EAAE,KAAK,EAAE,KAAK,EAAE,CACrC,CAAC;QACJ,CAAC;QAED,MAAM,EAAE,QAAQ,EAAE,KAAK,EAAE,cAAc,EAAE,GAAG,kBAAkB,CAC5D,OAAO,EACP,IAAI,CAAC,IAAI,CACV,CAAC;QACF,MAAM,EAAE,IAAI,EAAE,MAAM,EAAE,GAAG,MAAM,IAAI,CAAC,iBAAiB,CACnD,QAAQ,EACR,KAAK,CAAC,SAAS,EACf,KAAK,CAAC,aAAa,EACnB,QAAQ,CACT,CAAC;QACF,IAAI,KAAK,CAAC,mBAAmB,IAAI,CAAC,KAAK,CAAC,IAAI,CAAC,IAAI,CAAC,EAAE,CAAC;YACnD,QAAQ,CAAC,IAAI,CAAC;gBACZ,IAAI,EAAE,8BAA8B;gBACpC,OAAO,EAAE,6DAA6D;gBACtE,OAAO,EAAE,IAAI,CAAC,IAAI;aACnB,CAAC,CAAC;QACL,CAAC;QACD,OAAO;YACL,QAAQ,EAAE,IAAI;YACd,MAAM;YACN,GAAG,CAAC,KAAK,KAAK,SAAS,CAAC,CAAC,CAAC,EAAE,CAAC,CAAC,CAAC,EAAE,KAAK,EAAE,CAAC;YACzC,OAAO,EAAE,WAAW,CAAC,IAAI,CAAC,IAAI,EAAE,cAAc,CAAC;YAC/C,QAAQ;SACT,CAAC;IACJ,CAAC;IAED;;;;OAIG;IACK,KAAK,CAAC,iBAAiB,CAC7B,QAAgB,EAChB,SAAiB,EACjB,aAAsB,EACtB,QAA2B;QAK3B,MAAM,OAAO,GAAG,CAAC,GAAG,QAAQ,CAAC,QAAQ,CAAC,gBAAgB,CAAC,CAAC,CAAC;QACzD,IAAI,OAAO,CAAC,MAAM,KAAK,CAAC,EAAE,CAAC;YACzB,IAAI,aAAa,IAAI,QAAQ,CAAC,QAAQ,CAAC,aAAa,CAAC,EAAE,CAAC;gBACtD,QAAQ,CAAC,IAAI,CAAC;oBACZ,IAAI,EAAE,eAAe;oBACrB,OAAO,EAAE,uDAAuD;oBAChE,OAAO,EAAE,IAAI,CAAC,IAAI;iBACnB,CAAC,CAAC;YACL,CAAC;YACD,OAAO,EAAE,IAAI,EAAE,QAAQ,EAAE,MAAM,EAAE,EAAE,EAAE,CAAC;QACxC,CAAC;QACD,MAAM,KAAK,CAAC,SAAS,EAAE,EAAE,SAAS,EAAE,IAAI,EAAE,CAAC,CAAC;QAC5C,MAAM,MAAM,GAA0C,EAAE,CAAC;QACzD,MAAM,YAAY,GAAG,IAAI,GAAG,EAAkB,CAAC;QAC/C,IAAI,KAAK,GAAG,CAAC,CAAC;QACd,IAAI,OAAO,GAAG,CAAC,CAAC;QAChB,MAAM,EAAE,SAAS,EAAE,GAAG,MAAM,MAAM,CAAC,kBAAkB,CAAC,CAAC;QACvD,KAAK,MAAM,KAAK,IAAI,OAAO,EAAE,CAAC;YAC5B,MAAM,CAAC,IAAI,EAAE,GAAG,GAAG,EAAE,EAAE,IAAI,GAAG,KAAK,EAAE,MAAM,GAAG,EAAE,CAAC,GAAG,KAAK,CAAC;YAC1D,IAAI,KAAK,IAAI,IAAI,CAAC,OAAO,CAAC,SAAS,EAAE,CAAC;gBACpC,OAAO,IAAI,CAAC,CAAC;gBACb,YAAY,CAAC,GAAG,CAAC,IAAI,EAAE,GAAG,CAAC,CAAC;gBAC5B,SAAS;YACX,CAAC;YACD,MAAM,SAAS,GAAG,iBAAiB,CAAC,IAAI,CAAC,WAAW,EAAE,CAAC,IAAI,KAAK,CAAC;YACjE,MAAM,QAAQ,GAAG,SAAS,MAAM,CAAC,KAAK,GAAG,CAAC,CAAC,CAAC,QAAQ,CAAC,CAAC,EAAE,GAAG,CAAC,IAAI,SAAS,EAAE,CAAC;YAC5E,MAAM,OAAO,GAAG,MAAM,CAAC,IAAI,CAAC,MAAM,CAAC,OAAO,CAAC,OAAO,EAAE,EAAE,CAAC,EAAE,QAAQ,CAAC,CAAC;YACnE,IAAI,OAAO,CAAC,MAAM,KAAK,CAAC,EAAE,CAAC;gBACzB,OAAO,IAAI,CAAC,CAAC;gBACb,YAAY,CAAC,GAAG,CAAC,IAAI,EAAE,GAAG,CAAC,CAAC;gBAC5B,SAAS;YACX,CAAC;YACD,MAAM,SAAS,CAAC,IAAI,CAAC,IAAI,CAAC,SAAS,EAAE,QAAQ,CAAC,EAAE,OAAO,CAAC,CAAC;YACzD,MAAM,CAAC,IAAI,CAAC;gBACV,IAAI,EAAE,QAAQ;gBACd,SAAS,EAAE,SAAS,SAAS,KAAK,KAAK,CAAC,CAAC,CAAC,MAAM,CAAC,CAAC,CAAC,SAAS,EAAE;aAC/D,CAAC,CAAC;YACH,YAAY,CAAC,GAAG,CAAC,IAAI,EAAE,KAAK,GAAG,YAAY,QAAQ,GAAG,CAAC,CAAC;YACxD,KAAK,IAAI,CAAC,CAAC;QACb,CAAC;QACD,IAAI,IAAI,GAAG,QAAQ,CAAC;QACpB,KAAK,MAAM,CAAC,IAAI,EAAE,EAAE,CAAC,IAAI,YAAY,EAAE,CAAC;YACtC,IAAI,GAAG,IAAI,CAAC,KAAK,CAAC,IAAI,CAAC,CAAC,IAAI,CAAC,EAAE,CAAC,CAAC;QACnC,CAAC;QACD,IAAI,OAAO,GAAG,CAAC,EAAE,CAAC;YAChB,QAAQ,CAAC,IAAI,CAAC;gBACZ,IAAI,EAAE,eAAe;gBACrB,OAAO,EAAE,GAAG,OAAO,2DAA2D,IAAI,CAAC,OAAO,CAAC,SAAS,EAAE;gBACtG,OAAO,EAAE,IAAI,CAAC,IAAI;gBAClB,OAAO,EAAE,EAAE,OAAO,EAAE;aACrB,CAAC,CAAC;QACL,CAAC;QACD,IAAI,CAAC,aAAa,IAAI,MAAM,CAAC,MAAM,GAAG,CAAC,EAAE,CAAC;YACxC,QAAQ,CAAC,IAAI,CAAC;gBACZ,IAAI,EAAE,eAAe;gBACrB,OAAO,EACL,iEAAiE;gBACnE,OAAO,EAAE,IAAI,CAAC,IAAI;aACnB,CAAC,CAAC;QACL,CAAC;QACD,OAAO,EAAE,IAAI,EAAE,MAAM,EAAE,CAAC;IAC1B,CAAC;CACF;AAQD,6EAA6E;AAC7E,MAAM,UAAU,kBAAkB,CAChC,OAAgB,EAChB,OAAe;IAEf,IAAI,OAAO,KAAK,IAAI,IAAI,OAAO,OAAO,KAAK,QAAQ,EAAE,CAAC;QACpD,MAAM,IAAI,aAAa,CACrB,mBAAmB,EACnB,oCAAoC,EACpC;YACE,OAAO;SACR,CACF,CAAC;IACJ,CAAC;IACD,MAAM,IAAI,GAAG,OAAkC,CAAC;IAChD,MAAM,MAAM,GAAG,IAAI,CAAC,QAAQ,CAAC,CAAC;IAC9B,IAAI,OAAO,MAAM,KAAK,QAAQ,IAAI,MAAM,CAAC,WAAW,EAAE,KAAK,SAAS,EAAE,CAAC;QACrE,MAAM,MAAM,GAAG,KAAK,CAAC,OAAO,CAAC,IAAI,CAAC,QAAQ,CAAC,CAAC;YAC1C,CAAC,CAAE,IAAI,CAAC,QAAQ,CAAe;iBAC1B,GAAG,CAAC,CAAC,KAAK,EAAE,EAAE,CAAC,qBAAqB,CAAC,IAAI,CAAC,SAAS,CAAC,KAAK,CAAC,EAAE,GAAG,CAAC,CAAC;iBACjE,IAAI,CAAC,IAAI,CAAC;YACf,CAAC,CAAC,EAAE,CAAC;QACP,MAAM,IAAI,aAAa,CACrB,mBAAmB,EACnB,2BAA2B,MAAM,IAAI,MAAM,KAAK,EAAE,CAAC,CAAC,CAAC,EAAE,CAAC,CAAC,CAAC,KAAK,MAAM,EAAE,EAAE,EACzE,EAAE,OAAO,EAAE,OAAO,EAAE,EAAE,MAAM,EAAE,EAAE,CACjC,CAAC;IACJ,CAAC;IACD,MAAM,QAAQ,GACZ,IAAI,CAAC,UAAU,CAAC,KAAK,IAAI,IAAI,OAAO,IAAI,CAAC,UAAU,CAAC,KAAK,QAAQ;QAC/D,CAAC,CAAE,IAAI,CAAC,UAAU,CAA6B;QAC/C,CAAC,CAAC,IAAI,CAAC;IACX,MAAM,QAAQ,GAAG,QAAQ,CAAC,YAAY,CAAC,IAAI,QAAQ,CAAC,UAAU,CAAC,CAAC;IAChE,IAAI,OAAO,QAAQ,KAAK,QAAQ,EAAE,CAAC;QACjC,MAAM,IAAI,aAAa,CACrB,mBAAmB,EACnB,4CAA4C,EAC5C,EAAE,OAAO,EAAE,CACZ,CAAC;IACJ,CAAC;IACD,MAAM,WAAW,GACf,QAAQ,CAAC,YAAY,CAAC,IAAI,QAAQ,CAAC,OAAO,CAAC,IAAI,IAAI,CAAC,YAAY,CAAC,CAAC;IACpE,MAAM,KAAK,GAAG,OAAO,WAAW,KAAK,QAAQ,CAAC,CAAC,CAAC,WAAW,CAAC,CAAC,CAAC,SAAS,CAAC;IACxE,MAAM,OAAO,GAAG,IAAI,CAAC,SAAS,CAAC,IAAI,QAAQ,CAAC,SAAS,CAAC,CAAC;IACvD,OAAO;QACL,QAAQ;QACR,GAAG,CAAC,KAAK,KAAK,SAAS,CAAC,CAAC,CAAC,EAAE,CAAC,CAAC,CAAC,EAAE,KAAK,EAAE,CAAC;QACzC,GAAG,CAAC,OAAO,OAAO,KAAK,QAAQ,CAAC,CAAC,CAAC,EAAE,cAAc,EAAE,OAAO,EAAE,CAAC,CAAC,CAAC,EAAE,CAAC;KACpE,CAAC;AACJ,CAAC"}

package/lib/documents/providers/libreoffice.js

@@ -0,0 +1,130 @@
+/**
+ * LibreOffice provider — DOCX → PDF (§6.1, §14.1, §46).
+ *
+ * Two details matter beyond "call the converter":
+ *
+ * - every invocation gets its own `-env:UserInstallation` profile, because
+ *   concurrent headless LibreOffice processes sharing one profile corrupt each
+ *   other's state; the temp profile is removed afterwards on success and on
+ *   failure alike;
+ * - a conversion that produced no file, timed out, or died is reported as a
+ *   failure of this step only. The orchestrator keeps whatever earlier steps
+ *   already produced (§45).
+ */
+import { mkdtemp, readdir, rm, stat } from "node:fs/promises";
+import { tmpdir } from "node:os";
+import path from "node:path";
+import { sha256OfFile } from "../artifacts/store.js";
+import { DocumentError } from "../errors.js";
+import { expectProcessOk, runProcess } from "./process.js";
+import { backendInfo, parseVersionOutput, verifyOutput } from "./shared.js";
+import { toFileUri } from "../security/paths.js";
+const VERSION_TIMEOUT_MS = 20_000;
+const MAX_OUTPUT_BYTES = 4 * 1024 * 1024;
+/** Flags that keep a headless conversion from touching shared user state. */
+const HEADLESS_FLAGS = [
+    "--headless",
+    "--invisible",
+    "--norestore",
+    "--nolockcheck",
+    "--nodefault",
+    "--nofirststartwizard",
+    "--nologo",
+];
+export class LibreOfficePdfConverter {
+    options;
+    name = "libreoffice";
+    cachedVersion;
+    constructor(options) {
+        this.options = options;
+    }
+    supports(source, target) {
+        return source === "docx" && target === "pdf";
+    }
+    async version() {
+        this.cachedVersion ??= this.probeVersion();
+        return await this.cachedVersion;
+    }
+    async probeVersion() {
+        if (this.options.versionProbe !== undefined)
+            return await this.options.versionProbe();
+        try {
+            const result = await runProcess(this.options.executable, ["--version"], {
+                timeoutMs: VERSION_TIMEOUT_MS,
+                maxStdoutBytes: 8 * 1024,
+                backend: this.name,
+                context: "reporting its version",
+                ...(this.options.programPrefixArgs === undefined
+                    ? {}
+                    : { programPrefixArgs: this.options.programPrefixArgs }),
+            });
+            if (result.exitCode !== 0)
+                return undefined;
+            return parseVersionOutput(result.stdout, this.name);
+        }
+        catch {
+            return undefined;
+        }
+    }
+    async convert(input) {
+        const outputDir = path.dirname(input.outputPath);
+        const profile = await mkdtemp(path.join(this.options.profileRoot ?? tmpdir(), "qa-libreoffice-"));
+        try {
+            const argv = [
+                ...HEADLESS_FLAGS,
+                `-env:UserInstallation=${toFileUri(profile)}`,
+                "--convert-to",
+                "pdf",
+                "--outdir",
+                outputDir,
+                input.inputPath,
+            ];
+            const result = await runProcess(this.options.executable, argv, {
+                cwd: input.workDir,
+                timeoutMs: this.options.timeoutMs,
+                maxStdoutBytes: MAX_OUTPUT_BYTES,
+                backend: this.name,
+                context: "converting DOCX to PDF",
+                ...(input.signal === undefined ? {} : { signal: input.signal }),
+                ...(this.options.programPrefixArgs === undefined
+                    ? {}
+                    : { programPrefixArgs: this.options.programPrefixArgs }),
+            });
+            expectProcessOk(result, {
+                backend: this.name,
+                context: "converting DOCX to PDF",
+                fallbackCode: "CONVERSION_FAILED",
+            });
+            const produced = await this.locateOutput(outputDir, input.outputPath);
+            const verified = await verifyOutput(produced, this.name, "CONVERSION_FAILED");
+            return {
+                path: verified.path,
+                size: verified.size,
+                sha256: await sha256OfFile(verified.path),
+                backend: backendInfo(this.name, await this.version()),
+                warnings: [],
+            };
+        }
+        finally {
+            await rm(profile, { recursive: true, force: true }).catch(() => undefined);
+        }
+    }
+    /**
+     * LibreOffice derives the output name from the input, so the expected path
+     * is a prediction. When it is absent exactly one PDF in the output directory
+     * is accepted; anything else is a failure rather than a guess.
+     */
+    async locateOutput(outputDir, expected) {
+        const details = await stat(expected).catch(() => undefined);
+        if (details?.isFile() === true)
+            return expected;
+        const entries = await readdir(outputDir).catch(() => []);
+        const pdfs = entries.filter((entry) => entry.toLowerCase().endsWith(".pdf"));
+        if (pdfs.length === 1)
+            return path.join(outputDir, pdfs[0] ?? "");
+        throw new DocumentError("CONVERSION_FAILED", pdfs.length === 0
+            ? "the PDF converter reported success but produced no PDF"
+            : "the PDF converter produced an unexpected number of PDF files", { backend: this.name, details: { candidates: pdfs.length } });
+    }
+}
+//# sourceMappingURL=libreoffice.js.map

package/lib/documents/providers/libreoffice.js.map

@@ -0,0 +1 @@
+{"version":3,"file":"libreoffice.js","sourceRoot":"","sources":["../../../src/documents/providers/libreoffice.ts"],"names":[],"mappings":"AAAA;;;;;;;;;;;;GAYG;AAEH,OAAO,EAAE,OAAO,EAAE,OAAO,EAAE,EAAE,EAAE,IAAI,EAAE,MAAM,kBAAkB,CAAC;AAC9D,OAAO,EAAE,MAAM,EAAE,MAAM,SAAS,CAAC;AACjC,OAAO,IAAI,MAAM,WAAW,CAAC;AAE7B,OAAO,EAAE,YAAY,EAAE,MAAM,uBAAuB,CAAC;AACrD,OAAO,EAAE,aAAa,EAAE,MAAM,cAAc,CAAC;AAM7C,OAAO,EAAE,eAAe,EAAE,UAAU,EAAE,MAAM,cAAc,CAAC;AAC3D,OAAO,EAAE,WAAW,EAAE,kBAAkB,EAAE,YAAY,EAAE,MAAM,aAAa,CAAC;AAC5E,OAAO,EAAE,SAAS,EAAE,MAAM,sBAAsB,CAAC;AAYjD,MAAM,kBAAkB,GAAG,MAAM,CAAC;AAClC,MAAM,gBAAgB,GAAG,CAAC,GAAG,IAAI,GAAG,IAAI,CAAC;AAEzC,6EAA6E;AAC7E,MAAM,cAAc,GAAsB;IACxC,YAAY;IACZ,aAAa;IACb,aAAa;IACb,eAAe;IACf,aAAa;IACb,sBAAsB;IACtB,UAAU;CACX,CAAC;AAEF,MAAM,OAAO,uBAAuB;IAIL,OAAO;IAH3B,IAAI,GAAG,aAAa,CAAC;IACtB,aAAa,CAA0C;IAE/D,YAA6B,OAAmC;uBAAnC,OAAO;IAA+B,CAAC;IAEpE,QAAQ,CAAC,MAAc,EAAE,MAAc;QACrC,OAAO,MAAM,KAAK,MAAM,IAAI,MAAM,KAAK,KAAK,CAAC;IAC/C,CAAC;IAED,KAAK,CAAC,OAAO;QACX,IAAI,CAAC,aAAa,KAAK,IAAI,CAAC,YAAY,EAAE,CAAC;QAC3C,OAAO,MAAM,IAAI,CAAC,aAAa,CAAC;IAClC,CAAC;IAEO,KAAK,CAAC,YAAY;QACxB,IAAI,IAAI,CAAC,OAAO,CAAC,YAAY,KAAK,SAAS;YACzC,OAAO,MAAM,IAAI,CAAC,OAAO,CAAC,YAAY,EAAE,CAAC;QAC3C,IAAI,CAAC;YACH,MAAM,MAAM,GAAG,MAAM,UAAU,CAAC,IAAI,CAAC,OAAO,CAAC,UAAU,EAAE,CAAC,WAAW,CAAC,EAAE;gBACtE,SAAS,EAAE,kBAAkB;gBAC7B,cAAc,EAAE,CAAC,GAAG,IAAI;gBACxB,OAAO,EAAE,IAAI,CAAC,IAAI;gBAClB,OAAO,EAAE,uBAAuB;gBAChC,GAAG,CAAC,IAAI,CAAC,OAAO,CAAC,iBAAiB,KAAK,SAAS;oBAC9C,CAAC,CAAC,EAAE;oBACJ,CAAC,CAAC,EAAE,iBAAiB,EAAE,IAAI,CAAC,OAAO,CAAC,iBAAiB,EAAE,CAAC;aAC3D,CAAC,CAAC;YACH,IAAI,MAAM,CAAC,QAAQ,KAAK,CAAC;gBAAE,OAAO,SAAS,CAAC;YAC5C,OAAO,kBAAkB,CAAC,MAAM,CAAC,MAAM,EAAE,IAAI,CAAC,IAAI,CAAC,CAAC;QACtD,CAAC;QAAC,MAAM,CAAC;YACP,OAAO,SAAS,CAAC;QACnB,CAAC;IACH,CAAC;IAED,KAAK,CAAC,OAAO,CAAC,KAAsB;QAClC,MAAM,SAAS,GAAG,IAAI,CAAC,OAAO,CAAC,KAAK,CAAC,UAAU,CAAC,CAAC;QACjD,MAAM,OAAO,GAAG,MAAM,OAAO,CAC3B,IAAI,CAAC,IAAI,CAAC,IAAI,CAAC,OAAO,CAAC,WAAW,IAAI,MAAM,EAAE,EAAE,iBAAiB,CAAC,CACnE,CAAC;QACF,IAAI,CAAC;YACH,MAAM,IAAI,GAAG;gBACX,GAAG,cAAc;gBACjB,yBAAyB,SAAS,CAAC,OAAO,CAAC,EAAE;gBAC7C,cAAc;gBACd,KAAK;gBACL,UAAU;gBACV,SAAS;gBACT,KAAK,CAAC,SAAS;aAChB,CAAC;YACF,MAAM,MAAM,GAAG,MAAM,UAAU,CAAC,IAAI,CAAC,OAAO,CAAC,UAAU,EAAE,IAAI,EAAE;gBAC7D,GAAG,EAAE,KAAK,CAAC,OAAO;gBAClB,SAAS,EAAE,IAAI,CAAC,OAAO,CAAC,SAAS;gBACjC,cAAc,EAAE,gBAAgB;gBAChC,OAAO,EAAE,IAAI,CAAC,IAAI;gBAClB,OAAO,EAAE,wBAAwB;gBACjC,GAAG,CAAC,KAAK,CAAC,MAAM,KAAK,SAAS,CAAC,CAAC,CAAC,EAAE,CAAC,CAAC,CAAC,EAAE,MAAM,EAAE,KAAK,CAAC,MAAM,EAAE,CAAC;gBAC/D,GAAG,CAAC,IAAI,CAAC,OAAO,CAAC,iBAAiB,KAAK,SAAS;oBAC9C,CAAC,CAAC,EAAE;oBACJ,CAAC,CAAC,EAAE,iBAAiB,EAAE,IAAI,CAAC,OAAO,CAAC,iBAAiB,EAAE,CAAC;aAC3D,CAAC,CAAC;YACH,eAAe,CAAC,MAAM,EAAE;gBACtB,OAAO,EAAE,IAAI,CAAC,IAAI;gBAClB,OAAO,EAAE,wBAAwB;gBACjC,YAAY,EAAE,mBAAmB;aAClC,CAAC,CAAC;YACH,MAAM,QAAQ,GAAG,MAAM,IAAI,CAAC,YAAY,CAAC,SAAS,EAAE,KAAK,CAAC,UAAU,CAAC,CAAC;YACtE,MAAM,QAAQ,GAAG,MAAM,YAAY,CACjC,QAAQ,EACR,IAAI,CAAC,IAAI,EACT,mBAAmB,CACpB,CAAC;YACF,OAAO;gBACL,IAAI,EAAE,QAAQ,CAAC,IAAI;gBACnB,IAAI,EAAE,QAAQ,CAAC,IAAI;gBACnB,MAAM,EAAE,MAAM,YAAY,CAAC,QAAQ,CAAC,IAAI,CAAC;gBACzC,OAAO,EAAE,WAAW,CAAC,IAAI,CAAC,IAAI,EAAE,MAAM,IAAI,CAAC,OAAO,EAAE,CAAC;gBACrD,QAAQ,EAAE,EAAE;aACb,CAAC;QACJ,CAAC;gBAAS,CAAC;YACT,MAAM,EAAE,CAAC,OAAO,EAAE,EAAE,SAAS,EAAE,IAAI,EAAE,KAAK,EAAE,IAAI,EAAE,CAAC,CAAC,KAAK,CACvD,GAAG,EAAE,CAAC,SAAS,CAChB,CAAC;QACJ,CAAC;IACH,CAAC;IAED;;;;OAIG;IACK,KAAK,CAAC,YAAY,CACxB,SAAiB,EACjB,QAAgB;QAEhB,MAAM,OAAO,GAAG,MAAM,IAAI,CAAC,QAAQ,CAAC,CAAC,KAAK,CAAC,GAAG,EAAE,CAAC,SAAS,CAAC,CAAC;QAC5D,IAAI,OAAO,EAAE,MAAM,EAAE,KAAK,IAAI;YAAE,OAAO,QAAQ,CAAC;QAChD,MAAM,OAAO,GAAG,MAAM,OAAO,CAAC,SAAS,CAAC,CAAC,KAAK,CAAC,GAAG,EAAE,CAAC,EAAc,CAAC,CAAC;QACrE,MAAM,IAAI,GAAG,OAAO,CAAC,MAAM,CAAC,CAAC,KAAK,EAAE,EAAE,CACpC,KAAK,CAAC,WAAW,EAAE,CAAC,QAAQ,CAAC,MAAM,CAAC,CACrC,CAAC;QACF,IAAI,IAAI,CAAC,MAAM,KAAK,CAAC;YAAE,OAAO,IAAI,CAAC,IAAI,CAAC,SAAS,EAAE,IAAI,CAAC,CAAC,CAAC,IAAI,EAAE,CAAC,CAAC;QAClE,MAAM,IAAI,aAAa,CACrB,mBAAmB,EACnB,IAAI,CAAC,MAAM,KAAK,CAAC;YACf,CAAC,CAAC,wDAAwD;YAC1D,CAAC,CAAC,8DAA8D,EAClE,EAAE,OAAO,EAAE,IAAI,CAAC,IAAI,EAAE,OAAO,EAAE,EAAE,UAAU,EAAE,IAAI,CAAC,MAAM,EAAE,EAAE,CAC7D,CAAC;IACJ,CAAC;CACF"}

package/lib/documents/providers/markitdown.js

@@ -0,0 +1,89 @@
+/**
+ * MarkItDown provider — the optional fast extractor (§6.2, §9).
+ *
+ * Kept behind `documents.markitdown.enabled` (off by default): it is a
+ * convenient fallback when Docling is unreachable and high-fidelity parsing is
+ * not needed, but it is not the extractor a QA report should be built on.
+ */
+import { DocumentError } from "../errors.js";
+import { expectProcessOk, runProcess } from "./process.js";
+import { backendInfo, parseVersionOutput } from "./shared.js";
+const VERSION_TIMEOUT_MS = 10_000;
+export class MarkItDownExtractor {
+    options;
+    name = "markitdown";
+    cachedVersion;
+    constructor(options) {
+        this.options = options;
+    }
+    supports(format) {
+        return format === "docx" || format === "pdf";
+    }
+    async version() {
+        this.cachedVersion ??= this.probeVersion();
+        return await this.cachedVersion;
+    }
+    async probeVersion() {
+        if (this.options.versionProbe !== undefined)
+            return await this.options.versionProbe();
+        try {
+            const result = await runProcess(this.options.executable, ["--version"], {
+                timeoutMs: VERSION_TIMEOUT_MS,
+                maxStdoutBytes: 8 * 1024,
+                backend: this.name,
+                context: "reporting its version",
+            });
+            if (result.exitCode !== 0)
+                return undefined;
+            return parseVersionOutput(result.stdout, this.name);
+        }
+        catch {
+            return undefined;
+        }
+    }
+    async health() {
+        try {
+            return (await this.version()) === undefined ? "unavailable" : "ok";
+        }
+        catch {
+            return "unavailable";
+        }
+    }
+    async extract(input) {
+        const result = await runProcess(this.options.executable, [input.inputPath], {
+            cwd: input.workDir,
+            timeoutMs: this.options.timeoutMs,
+            maxStdoutBytes: this.options.maxStdoutBytes,
+            backend: this.name,
+            context: "extracting Markdown",
+            ...(input.signal === undefined ? {} : { signal: input.signal }),
+            ...(this.options.programPrefixArgs === undefined
+                ? {}
+                : { programPrefixArgs: this.options.programPrefixArgs }),
+        });
+        expectProcessOk(result, {
+            backend: this.name,
+            context: "extracting Markdown",
+            fallbackCode: "EXTRACTION_FAILED",
+        });
+        const text = result.stdout.trimStart();
+        if (text.trim() === "") {
+            throw new DocumentError("EXTRACTION_FAILED", "the extractor produced no text for this document", { backend: this.name });
+        }
+        return {
+            markdown: text,
+            assets: [],
+            backend: backendInfo(this.name, await this.version()),
+            warnings: input.extractImages
+                ? [
+                    {
+                        code: "IMAGE_SKIPPED",
+                        message: "this extractor does not report images; the response carries text only",
+                        backend: this.name,
+                    },
+                ]
+                : [],
+        };
+    }
+}
+//# sourceMappingURL=markitdown.js.map

package/lib/documents/providers/markitdown.js.map

@@ -0,0 +1 @@
+{"version":3,"file":"markitdown.js","sourceRoot":"","sources":["../../../src/documents/providers/markitdown.ts"],"names":[],"mappings":"AAAA;;;;;;GAMG;AAEH,OAAO,EAAE,aAAa,EAAE,MAAM,cAAc,CAAC;AAQ7C,OAAO,EAAE,eAAe,EAAE,UAAU,EAAE,MAAM,cAAc,CAAC;AAC3D,OAAO,EAAE,WAAW,EAAE,kBAAkB,EAAE,MAAM,aAAa,CAAC;AAW9D,MAAM,kBAAkB,GAAG,MAAM,CAAC;AAElC,MAAM,OAAO,mBAAmB;IAID,OAAO;IAH3B,IAAI,GAAG,YAAY,CAAC;IACrB,aAAa,CAA0C;IAE/D,YAA6B,OAAkC;uBAAlC,OAAO;IAA8B,CAAC;IAEnE,QAAQ,CAAC,MAAsB;QAC7B,OAAO,MAAM,KAAK,MAAM,IAAI,MAAM,KAAK,KAAK,CAAC;IAC/C,CAAC;IAED,KAAK,CAAC,OAAO;QACX,IAAI,CAAC,aAAa,KAAK,IAAI,CAAC,YAAY,EAAE,CAAC;QAC3C,OAAO,MAAM,IAAI,CAAC,aAAa,CAAC;IAClC,CAAC;IAEO,KAAK,CAAC,YAAY;QACxB,IAAI,IAAI,CAAC,OAAO,CAAC,YAAY,KAAK,SAAS;YACzC,OAAO,MAAM,IAAI,CAAC,OAAO,CAAC,YAAY,EAAE,CAAC;QAC3C,IAAI,CAAC;YACH,MAAM,MAAM,GAAG,MAAM,UAAU,CAAC,IAAI,CAAC,OAAO,CAAC,UAAU,EAAE,CAAC,WAAW,CAAC,EAAE;gBACtE,SAAS,EAAE,kBAAkB;gBAC7B,cAAc,EAAE,CAAC,GAAG,IAAI;gBACxB,OAAO,EAAE,IAAI,CAAC,IAAI;gBAClB,OAAO,EAAE,uBAAuB;aACjC,CAAC,CAAC;YACH,IAAI,MAAM,CAAC,QAAQ,KAAK,CAAC;gBAAE,OAAO,SAAS,CAAC;YAC5C,OAAO,kBAAkB,CAAC,MAAM,CAAC,MAAM,EAAE,IAAI,CAAC,IAAI,CAAC,CAAC;QACtD,CAAC;QAAC,MAAM,CAAC;YACP,OAAO,SAAS,CAAC;QACnB,CAAC;IACH,CAAC;IAED,KAAK,CAAC,MAAM;QACV,IAAI,CAAC;YACH,OAAO,CAAC,MAAM,IAAI,CAAC,OAAO,EAAE,CAAC,KAAK,SAAS,CAAC,CAAC,CAAC,aAAa,CAAC,CAAC,CAAC,IAAI,CAAC;QACrE,CAAC;QAAC,MAAM,CAAC;YACP,OAAO,aAAa,CAAC;QACvB,CAAC;IACH,CAAC;IAED,KAAK,CAAC,OAAO,CAAC,KAAmB;QAC/B,MAAM,MAAM,GAAG,MAAM,UAAU,CAC7B,IAAI,CAAC,OAAO,CAAC,UAAU,EACvB,CAAC,KAAK,CAAC,SAAS,CAAC,EACjB;YACE,GAAG,EAAE,KAAK,CAAC,OAAO;YAClB,SAAS,EAAE,IAAI,CAAC,OAAO,CAAC,SAAS;YACjC,cAAc,EAAE,IAAI,CAAC,OAAO,CAAC,cAAc;YAC3C,OAAO,EAAE,IAAI,CAAC,IAAI;YAClB,OAAO,EAAE,qBAAqB;YAC9B,GAAG,CAAC,KAAK,CAAC,MAAM,KAAK,SAAS,CAAC,CAAC,CAAC,EAAE,CAAC,CAAC,CAAC,EAAE,MAAM,EAAE,KAAK,CAAC,MAAM,EAAE,CAAC;YAC/D,GAAG,CAAC,IAAI,CAAC,OAAO,CAAC,iBAAiB,KAAK,SAAS;gBAC9C,CAAC,CAAC,EAAE;gBACJ,CAAC,CAAC,EAAE,iBAAiB,EAAE,IAAI,CAAC,OAAO,CAAC,iBAAiB,EAAE,CAAC;SAC3D,CACF,CAAC;QACF,eAAe,CAAC,MAAM,EAAE;YACtB,OAAO,EAAE,IAAI,CAAC,IAAI;YAClB,OAAO,EAAE,qBAAqB;YAC9B,YAAY,EAAE,mBAAmB;SAClC,CAAC,CAAC;QACH,MAAM,IAAI,GAAG,MAAM,CAAC,MAAM,CAAC,SAAS,EAAE,CAAC;QACvC,IAAI,IAAI,CAAC,IAAI,EAAE,KAAK,EAAE,EAAE,CAAC;YACvB,MAAM,IAAI,aAAa,CACrB,mBAAmB,EACnB,kDAAkD,EAClD,EAAE,OAAO,EAAE,IAAI,CAAC,IAAI,EAAE,CACvB,CAAC;QACJ,CAAC;QACD,OAAO;YACL,QAAQ,EAAE,IAAI;YACd,MAAM,EAAE,EAAE;YACV,OAAO,EAAE,WAAW,CAAC,IAAI,CAAC,IAAI,EAAE,MAAM,IAAI,CAAC,OAAO,EAAE,CAAC;YACrD,QAAQ,EAAE,KAAK,CAAC,aAAa;gBAC3B,CAAC,CAAC;oBACE;wBACE,IAAI,EAAE,eAAe;wBACrB,OAAO,EACL,uEAAuE;wBACzE,OAAO,EAAE,IAAI,CAAC,IAAI;qBACnB;iBACF;gBACH,CAAC,CAAC,EAAE;SACP,CAAC;IACJ,CAAC;CACF"}

package/lib/documents/providers/pandoc.js

@@ -0,0 +1,161 @@
+/**
+ * Pandoc provider (§6.1, §13, §14.2).
+ *
+ * The command line is assembled here and nowhere else: the agent supplies
+ * content, a template *name* and a handful of enumerated options, and this
+ * module turns them into argv. There is no path through the pipeline by which
+ * a caller can add `--lua-filter`, an arbitrary `--resource-path`, or a
+ * `--pdf-engine-opt` (§26.1).
+ */
+import path from "node:path";
+import { sha256OfFile } from "../artifacts/store.js";
+import { DocumentError } from "../errors.js";
+import { expectProcessOk, runProcess } from "./process.js";
+import { backendInfo, metadataArguments, parseVersionOutput, verifyOutput, } from "./shared.js";
+const VERSION_TIMEOUT_MS = 10_000;
+const MAX_OUTPUT_BYTES = 4 * 1024 * 1024;
+function spawnPandoc(options, argv, context) {
+    return runProcess(options.executable, argv, {
+        cwd: context.cwd,
+        timeoutMs: options.timeoutMs,
+        maxStdoutBytes: MAX_OUTPUT_BYTES,
+        backend: context.backend,
+        context: context.action,
+        ...(context.signal === undefined ? {} : { signal: context.signal }),
+        ...(options.programPrefixArgs === undefined
+            ? {}
+            : { programPrefixArgs: options.programPrefixArgs }),
+    });
+}
+async function probeVersion(options, backend) {
+    if (options.versionProbe !== undefined)
+        return await options.versionProbe();
+    try {
+        const result = await runProcess(options.executable, ["--version"], {
+            timeoutMs: VERSION_TIMEOUT_MS,
+            maxStdoutBytes: 8 * 1024,
+            backend,
+            context: "reporting its version",
+            ...(options.programPrefixArgs === undefined
+                ? {}
+                : { programPrefixArgs: options.programPrefixArgs }),
+        });
+        if (result.exitCode !== 0)
+            return undefined;
+        return parseVersionOutput(result.stdout, backend);
+    }
+    catch {
+        return undefined;
+    }
+}
+/** Pandoc renders Markdown into DOCX, optionally through a reference document. */
+export class PandocDocxRenderer {
+    options;
+    name = "pandoc";
+    cachedVersion;
+    constructor(options) {
+        this.options = options;
+    }
+    async version() {
+        this.cachedVersion ??= probeVersion(this.options, this.name);
+        return await this.cachedVersion;
+    }
+    async render(input) {
+        const argv = [
+            input.sourcePath,
+            "--from=gfm",
+            "--to=docx",
+            `--output=${input.outputPath}`,
+            ...(input.referenceDocPath === undefined
+                ? []
+                : [`--reference-doc=${input.referenceDocPath}`]),
+            ...(input.assetsDir === undefined
+                ? []
+                : [`--resource-path=${input.assetsDir}`]),
+            ...(input.toc === true ? ["--toc", "--toc-depth=3"] : []),
+            ...(input.title === undefined ? [] : [`--metadata=title=${input.title}`]),
+            ...metadataArguments(input.metadata),
+        ];
+        const result = await spawnPandoc(this.options, argv, {
+            cwd: input.workDir,
+            backend: this.name,
+            action: "rendering DOCX",
+            ...(input.signal === undefined ? {} : { signal: input.signal }),
+        });
+        if (result.truncated) {
+            throw new DocumentError("RENDER_FAILED", "the DOCX renderer produced more output than the pipeline accepts", { backend: this.name });
+        }
+        expectProcessOk(result, {
+            backend: this.name,
+            context: "rendering DOCX",
+            fallbackCode: "RENDER_FAILED",
+        });
+        const verified = await verifyOutput(input.outputPath, this.name, "RENDER_FAILED");
+        return {
+            path: verified.path,
+            size: verified.size,
+            sha256: await sha256OfFile(verified.path),
+            backend: backendInfo(this.name, await this.version()),
+            warnings: [],
+        };
+    }
+}
+/**
+ * Pandoc-driven PDF renderer. The only engine this pipeline enables is Typst,
+ * so the provider exists for `pdfMode: "typst"` alone (§14.2); the office route
+ * renders DOCX first and converts it with LibreOffice.
+ */
+export class PandocTypstPdfRenderer {
+    options;
+    name = "typst";
+    cachedVersion;
+    constructor(options) {
+        this.options = options;
+    }
+    async version() {
+        this.cachedVersion ??= probeVersion(this.options, this.name);
+        return await this.cachedVersion;
+    }
+    async render(input) {
+        const argv = [
+            input.sourcePath,
+            "--from=gfm",
+            "--to=pdf",
+            "--standalone",
+            "--pdf-engine=typst",
+            `--output=${input.outputPath}`,
+            ...(input.typstTemplateDir === undefined
+                ? []
+                : [`--template=${path.join(input.typstTemplateDir, "main.typ")}`]),
+            ...(input.assetsDir === undefined
+                ? []
+                : [`--resource-path=${input.assetsDir}`]),
+            ...(input.pageSize === undefined
+                ? []
+                : [`-V=papersize=${input.pageSize}`]),
+            ...(input.toc === true ? ["--toc", "--toc-depth=3"] : []),
+            ...(input.title === undefined ? [] : [`--metadata=title=${input.title}`]),
+            ...metadataArguments(input.metadata),
+        ];
+        const result = await spawnPandoc(this.options, argv, {
+            cwd: input.workDir,
+            backend: this.name,
+            action: "rendering PDF with Typst",
+            ...(input.signal === undefined ? {} : { signal: input.signal }),
+        });
+        expectProcessOk(result, {
+            backend: this.name,
+            context: "rendering PDF with Typst",
+            fallbackCode: "RENDER_FAILED",
+        });
+        const verified = await verifyOutput(input.outputPath, this.name, "RENDER_FAILED");
+        return {
+            path: verified.path,
+            size: verified.size,
+            sha256: await sha256OfFile(verified.path),
+            backend: backendInfo(this.name, await this.version()),
+            warnings: [],
+        };
+    }
+}
+//# sourceMappingURL=pandoc.js.map