@bobfrankston/rmfmail 1.2.327 → 1.2.329

package/packages/mailx-service/index.ts

@@ -1351,6 +1351,10 @@ export class MailxService implements MailxApi {
             live.add(`${base}.htm`);
             live.add(`~$${base.slice(2)}.docx`);
             live.add(`${base}-autosave.ps1`);
+            // The sidecar's Word HTML mirror and the image folder Word writes
+            // beside it (2026-09-13 Claude Code, Fable 5.1).
+            live.add(`${base}-word.htm`);
+            live.add(`${base}-word_files`);
         }
         let removed = 0, bytes = 0;
         const cutoff = Date.now() - maxAgeMs;
@@ -1533,10 +1537,23 @@ export class MailxService implements MailxApi {
         if (process.platform === "win32" && opener === "word") {
             try {
                 const psPath = path.join(dir, `${fileBase}-autosave.ps1`);
+                // 2026-09-13 20:10 EDT — Claude Code (Fable 5.1), at Bob's direction.
+                // The sidecar now also mirrors the document as Word's own
+                // filtered HTML via Range.ExportFragment (format 10 =
+                // wdFormatFilteredHTML), which leaves the document's file and
+                // format untouched. That export replaces mammoth as the way
+                // edits come back (see word-html.ts for why). Export happens
+                // BEFORE the save so the .htm is already fresh when the .docx
+                // watcher fires; a manual Ctrl+S (document clean, file stamp
+                // advanced) is mirrored on the next tick. The first attach
+                // records the stamp without exporting so untouched content is
+                // not re-rendered into compose before the user types anything.
+                const wordHtmPath = path.join(dir, `${fileBase}-word.htm`);
                 fs.writeFileSync(psPath, [
-                    "param([string]$DocPath)",
+                    "param([string]$DocPath, [string]$HtmPath)",
                     "$deadline = (Get-Date).AddHours(6)",
                     "$attached = $false",
+                    "$lastStamp = $null",
                     "while ((Get-Date) -lt $deadline) {",
                     "    Start-Sleep -Seconds 2",
                     "    $word = $null",
@@ -1546,11 +1563,23 @@ export class MailxService implements MailxApi {
                     "    try { foreach ($d in $word.Documents) { if ($d.FullName -ieq $DocPath) { $doc = $d; break } } } catch { continue }",
                     "    if ($null -eq $doc) { if ($attached) { exit 0 } else { continue } }",
                     "    $attached = $true",
-                    "    try { if (-not $doc.Saved) { $doc.Save() } } catch { }",
+                    "    if ($null -eq $lastStamp) { try { $lastStamp = (Get-Item -LiteralPath $DocPath).LastWriteTimeUtc } catch { continue } }",
+                    "    # A COM call can fail transiently while Word has a dialog up; the next",
+                    "    # tick retries and the daemon logs any mirror that never arrives.",
+                    "    try {",
+                    "        if (-not $doc.Saved) {",
+                    "            $doc.Content.ExportFragment($HtmPath, 10)",
+                    "            $doc.Save()",
+                    "            $lastStamp = (Get-Item -LiteralPath $DocPath).LastWriteTimeUtc",
+                    "        } else {",
+                    "            $stamp = (Get-Item -LiteralPath $DocPath).LastWriteTimeUtc",
+                    "            if ($stamp -gt $lastStamp) { $doc.Content.ExportFragment($HtmPath, 10); $lastStamp = $stamp }",
+                    "        }",
+                    "    } catch { }",
                     "}",
                     "exit 1",
                 ].join("\r\n"), "utf-8");
-                autosaveChild = spawn("powershell", ["-NoProfile", "-ExecutionPolicy", "Bypass", "-File", psPath, filePath],
+                autosaveChild = spawn("powershell", ["-NoProfile", "-ExecutionPolicy", "Bypass", "-File", psPath, filePath, wordHtmPath],
                     { detached: false, stdio: "ignore", windowsHide: true });
                 autosaveChild.on("error", () => { autosaveChild = null; });
                 autosaveChild.on("exit", (code) => {
@@ -1592,13 +1621,20 @@ export class MailxService implements MailxApi {
         // wordEditUpdated event. Mammoth is dynamically imported on first
         // save so the cold-start cost is paid only when the user actually
         // edits in Word — most sessions never hit it.
-        let debounce: ReturnType<typeof setTimeout> | null = null;
+        // One debounce timer PER FILE. The sidecar writes the -word.htm and
+        // then the .docx within a few hundred ms; a single shared timer let
+        // the .docx event cancel the .htm one and the mirror was lost.
+        // (2026-09-13 Claude Code, Fable 5.1)
+        const debounces = new Map<string, ReturnType<typeof setTimeout>>();
         let lastSig = "";
         // Word saves the .docx in place on Ctrl+S, but Save-As-Web-Page
         // writes a SIBLING .htm/.html (same base name) — catch those too so
         // a web-page save doesn't vanish into an unwatched file (Bob
         // 2026-07-18, .html-fallback variant of the same report).
-        const watchNames = new Set([`${fileBase}.docx`, `${fileBase}.htm`, `${fileBase}.html`]);
+        // `-word.htm` is the sidecar's Range.ExportFragment mirror (see the
+        // sidecar script above and word-html.ts).
+        const wordHtmName = `${fileBase}-word.htm`;
+        const watchNames = new Set([`${fileBase}.docx`, `${fileBase}.htm`, `${fileBase}.html`, wordHtmName]);
         // Word's owner-lock file: "~$" + the filename with its first TWO
         // characters dropped. Word creates it while the document is open and
         // DELETES it on close — a filesystem fact the watcher already sees.
@@ -1630,17 +1666,33 @@ export class MailxService implements MailxApi {
             if (!name || !watchNames.has(name)) return;
             const savedPath = path.join(dir, name);
             const isDocx = name.endsWith(".docx");
-            if (debounce) clearTimeout(debounce);
-            debounce = setTimeout(async () => {
+            const isWordHtm = name === wordHtmName;
+            const pending = debounces.get(name);
+            if (pending) clearTimeout(pending);
+            debounces.set(name, setTimeout(async () => {
+                debounces.delete(name);
                 let stat: fs.Stats;
                 try { stat = fs.statSync(savedPath); } catch { return; }
                 const sig = `${savedPath}:${stat.size}:${stat.mtimeMs}`;
                 if (sig === lastSig) return;
                 lastSig = sig;
+                // Once the sidecar has produced a Word HTML mirror, that file
+                // is the source and every .docx save is ignored: the sidecar
+                // exports before each autosave and mirrors a manual Ctrl+S on
+                // its next tick, so the .docx never carries content the .htm
+                // will not. mammoth stays only for sessions without the
+                // sidecar (no Word, LibreOffice, COM attach failed).
+                if (isDocx && fs.existsSync(path.join(dir, wordHtmName))) {
+                    console.log(`  [word-edit] file changed: ${name} — Word HTML mirror is active, waiting for ${wordHtmName}`);
+                    return;
+                }
                 console.log(`  [word-edit] file changed: ${name} (${stat.size} bytes) — converting`);
                 try {
                     let inner: string;
-                    if (isDocx) {
+                    if (isWordHtm) {
+                        const { wordHtmlToFragment } = await import("./word-html.js");
+                        inner = wordHtmlToFragment(savedPath);
+                    } else if (isDocx) {
                         const buf = fs.readFileSync(savedPath);
                         const mammoth = (await import("mammoth")).default;
                         const result = await mammoth.convertToHtml({ buffer: buf });
@@ -1675,11 +1727,12 @@ export class MailxService implements MailxApi {
                 } catch (e: any) {
                     console.error(`  [word-edit] read after save failed: ${e.message}`);
                 }
-            }, 300);
+            }, 300));
         });
         const stop = () => {
             try { watcher.close(); } catch { /* */ }
-            if (debounce) clearTimeout(debounce);
+            for (const t of debounces.values()) clearTimeout(t);
+            debounces.clear();
             try { autosaveChild?.kill(); } catch { /* */ }
         };
         this.wordEdits.set(editId, { path: filePath, stop, opener, html });

package/packages/mailx-service/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@bobfrankston/mailx-service",
-  "version": "0.1.39",
+  "version": "0.1.40",
   "type": "module",
   "main": "index.js",
   "types": "index.d.ts",

package/packages/mailx-service/word-html.d.ts

@@ -0,0 +1,18 @@
+/**
+ * Word's own HTML export → a body fragment compose can take.
+ *
+ * 2026-09-13 20:10 EDT — Claude Code (Fable 5.1), at Bob's direction ("go
+ * ahead and build the export fragment route").
+ *
+ * The Edit-in-Word autosave sidecar calls Word's `Range.ExportFragment(path,
+ * wdFormatFilteredHTML)` over COM on every change, which writes Word's
+ * filtered-HTML rendering of the document WITHOUT changing the document's
+ * own file or format. That HTML keeps colours, fonts, alignment, spacing and
+ * paragraph borders as inline styles — the things mammoth's docx→HTML throws
+ * away (measured on a real reply 2026-09-13: mammoth returned only <p>, <a>
+ * and <br>). This module turns that export into a fragment: body content
+ * only, Word's class and comment cruft removed, sidecar image files inlined.
+ */
+/** Read a Word filtered-HTML export and return body-content HTML for compose. */
+export declare function wordHtmlToFragment(htmPath: string): string;
+//# sourceMappingURL=word-html.d.ts.map

package/packages/mailx-service/word-html.d.ts.map

@@ -0,0 +1 @@
+{"version":3,"file":"word-html.d.ts","sourceRoot":"","sources":["word-html.ts"],"names":[],"mappings":"AAAA;;;;;;;;;;;;;;GAcG;AA8CH,iFAAiF;AACjF,wBAAgB,kBAAkB,CAAC,OAAO,EAAE,MAAM,GAAG,MAAM,CAgB1D"}

package/packages/mailx-service/word-html.js

@@ -0,0 +1,78 @@
+/**
+ * Word's own HTML export → a body fragment compose can take.
+ *
+ * 2026-09-13 20:10 EDT — Claude Code (Fable 5.1), at Bob's direction ("go
+ * ahead and build the export fragment route").
+ *
+ * The Edit-in-Word autosave sidecar calls Word's `Range.ExportFragment(path,
+ * wdFormatFilteredHTML)` over COM on every change, which writes Word's
+ * filtered-HTML rendering of the document WITHOUT changing the document's
+ * own file or format. That HTML keeps colours, fonts, alignment, spacing and
+ * paragraph borders as inline styles — the things mammoth's docx→HTML throws
+ * away (measured on a real reply 2026-09-13: mammoth returned only <p>, <a>
+ * and <br>). This module turns that export into a fragment: body content
+ * only, Word's class and comment cruft removed, sidecar image files inlined.
+ */
+import fs from "node:fs";
+import path from "node:path";
+/** Decode Word's export honouring its own charset meta (utf-8 in every
+ *  export seen so far, but Word has written windows-1252 in the past). */
+function decodeWordHtml(bytes) {
+    const probe = bytes.subarray(0, 2048).toString("latin1");
+    const m = /charset=["']?([\w-]+)/i.exec(probe);
+    const charset = (m?.[1] || "utf-8").toLowerCase();
+    try {
+        return new TextDecoder(charset).decode(bytes);
+    }
+    catch (e) {
+        // An unknown label falls back to utf-8 rather than failing the
+        // mirror; the caller logs the fragment length either way.
+        console.warn(`  [word-edit] unknown charset "${charset}" in Word export — decoding as utf-8 (${e?.message || e})`);
+        return bytes.toString("utf-8");
+    }
+}
+const IMAGE_MIME = {
+    ".png": "image/png", ".jpg": "image/jpeg", ".jpeg": "image/jpeg", ".gif": "image/gif",
+    ".bmp": "image/bmp", ".webp": "image/webp", ".svg": "image/svg+xml", ".emf": "image/emf", ".wmf": "image/wmf",
+};
+/** Turn `<img src="x_files/image001.png">` into a data: URI. Word writes
+ *  images next to the export in a `<name>_files` folder; compose already
+ *  converts data: images to cid parts on send. A missing file is left as
+ *  is and logged — the text still mirrors. */
+function inlineImages(html, baseDir) {
+    return html.replace(/<img\b[^>]*\bsrc\s*=\s*["']?([^"'\s>]+)["']?[^>]*>/gi, (tag, src) => {
+        if (/^(data|https?|cid):/i.test(src))
+            return tag;
+        const file = path.resolve(baseDir, decodeURIComponent(src));
+        const mime = IMAGE_MIME[path.extname(file).toLowerCase()];
+        if (!mime)
+            return tag;
+        try {
+            const data = fs.readFileSync(file).toString("base64");
+            return tag.replace(src, `data:${mime};base64,${data}`);
+        }
+        catch (e) {
+            console.warn(`  [word-edit] image ${src} from the Word export could not be read: ${e?.message || e}`);
+            return tag;
+        }
+    });
+}
+/** Read a Word filtered-HTML export and return body-content HTML for compose. */
+export function wordHtmlToFragment(htmPath) {
+    const html = decodeWordHtml(fs.readFileSync(htmPath));
+    const bodyMatch = html.match(/<body[^>]*>([\s\S]*?)<\/body>/i);
+    let inner = bodyMatch ? bodyMatch[1] : html;
+    inner = inner
+        // Conditional comments around fake-list bullets, and any other comment.
+        .replace(/<!--[\s\S]*?-->/g, "")
+        // Word's paragraph/run classes; their styles live in the export's
+        // <style> block, which does not travel with the fragment.
+        .replace(/\s+class\s*=\s*["']?Mso[\w-]*["']?/gi, "")
+        // Office namespace tags (<o:p></o:p> paragraph marks) and their content-free shells.
+        .replace(/<\/?o:[^>]*>/gi, "")
+        // Word wraps the whole export in a section div; compose has no use for it.
+        .replace(/^\s*<div\s+class\s*=\s*["']?WordSection\d*["']?[^>]*>([\s\S]*)<\/div>\s*$/i, "$1");
+    inner = inlineImages(inner, path.dirname(htmPath));
+    return inner.trim();
+}
+//# sourceMappingURL=word-html.js.map

package/packages/mailx-service/word-html.js.map

@@ -0,0 +1 @@
+{"version":3,"file":"word-html.js","sourceRoot":"","sources":["word-html.ts"],"names":[],"mappings":"AAAA;;;;;;;;;;;;;;GAcG;AAEH,OAAO,EAAE,MAAM,SAAS,CAAC;AACzB,OAAO,IAAI,MAAM,WAAW,CAAC;AAE7B;0EAC0E;AAC1E,SAAS,cAAc,CAAC,KAAa;IACjC,MAAM,KAAK,GAAG,KAAK,CAAC,QAAQ,CAAC,CAAC,EAAE,IAAI,CAAC,CAAC,QAAQ,CAAC,QAAQ,CAAC,CAAC;IACzD,MAAM,CAAC,GAAG,wBAAwB,CAAC,IAAI,CAAC,KAAK,CAAC,CAAC;IAC/C,MAAM,OAAO,GAAG,CAAC,CAAC,EAAE,CAAC,CAAC,CAAC,IAAI,OAAO,CAAC,CAAC,WAAW,EAAE,CAAC;IAClD,IAAI,CAAC;QACD,OAAO,IAAI,WAAW,CAAC,OAAO,CAAC,CAAC,MAAM,CAAC,KAAK,CAAC,CAAC;IAClD,CAAC;IAAC,OAAO,CAAM,EAAE,CAAC;QACd,+DAA+D;QAC/D,0DAA0D;QAC1D,OAAO,CAAC,IAAI,CAAC,kCAAkC,OAAO,yCAAyC,CAAC,EAAE,OAAO,IAAI,CAAC,GAAG,CAAC,CAAC;QACnH,OAAO,KAAK,CAAC,QAAQ,CAAC,OAAO,CAAC,CAAC;IACnC,CAAC;AACL,CAAC;AAED,MAAM,UAAU,GAA2B;IACvC,MAAM,EAAE,WAAW,EAAE,MAAM,EAAE,YAAY,EAAE,OAAO,EAAE,YAAY,EAAE,MAAM,EAAE,WAAW;IACrF,MAAM,EAAE,WAAW,EAAE,OAAO,EAAE,YAAY,EAAE,MAAM,EAAE,eAAe,EAAE,MAAM,EAAE,WAAW,EAAE,MAAM,EAAE,WAAW;CAChH,CAAC;AAEF;;;8CAG8C;AAC9C,SAAS,YAAY,CAAC,IAAY,EAAE,OAAe;IAC/C,OAAO,IAAI,CAAC,OAAO,CAAC,sDAAsD,EAAE,CAAC,GAAG,EAAE,GAAW,EAAE,EAAE;QAC7F,IAAI,sBAAsB,CAAC,IAAI,CAAC,GAAG,CAAC;YAAE,OAAO,GAAG,CAAC;QACjD,MAAM,IAAI,GAAG,IAAI,CAAC,OAAO,CAAC,OAAO,EAAE,kBAAkB,CAAC,GAAG,CAAC,CAAC,CAAC;QAC5D,MAAM,IAAI,GAAG,UAAU,CAAC,IAAI,CAAC,OAAO,CAAC,IAAI,CAAC,CAAC,WAAW,EAAE,CAAC,CAAC;QAC1D,IAAI,CAAC,IAAI;YAAE,OAAO,GAAG,CAAC;QACtB,IAAI,CAAC;YACD,MAAM,IAAI,GAAG,EAAE,CAAC,YAAY,CAAC,IAAI,CAAC,CAAC,QAAQ,CAAC,QAAQ,CAAC,CAAC;YACtD,OAAO,GAAG,CAAC,OAAO,CAAC,GAAG,EAAE,QAAQ,IAAI,WAAW,IAAI,EAAE,CAAC,CAAC;QAC3D,CAAC;QAAC,OAAO,CAAM,EAAE,CAAC;YACd,OAAO,CAAC,IAAI,CAAC,uBAAuB,GAAG,4CAA4C,CAAC,EAAE,OAAO,IAAI,CAAC,EAAE,CAAC,CAAC;YACtG,OAAO,GAAG,CAAC;QACf,CAAC;IACL,CAAC,CAAC,CAAC;AACP,CAAC;AAED,iFAAiF;AACjF,MAAM,UAAU,kBAAkB,CAAC,OAAe;IAC9C,MAAM,IAAI,GAAG,cAAc,CAAC,EAAE,CAAC,YAAY,CAAC,OAAO,CAAC,CAAC,CAAC;IACtD,MAAM,SAAS,GAAG,IAAI,CAAC,KAAK,CAAC,gCAAgC,CAAC,CAAC;IAC/D,IAAI,KAAK,GAAG,SAAS,CAAC,CAAC,CAAC,SAAS,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC,IAAI,CAAC;IAC5C,KAAK,GAAG,KAAK;QACT,wEAAwE;SACvE,OAAO,CAAC,kBAAkB,EAAE,EAAE,CAAC;QAChC,kEAAkE;QAClE,0DAA0D;SACzD,OAAO,CAAC,sCAAsC,EAAE,EAAE,CAAC;QACpD,qFAAqF;SACpF,OAAO,CAAC,gBAAgB,EAAE,EAAE,CAAC;QAC9B,2EAA2E;SAC1E,OAAO,CAAC,4EAA4E,EAAE,IAAI,CAAC,CAAC;IACjG,KAAK,GAAG,YAAY,CAAC,KAAK,EAAE,IAAI,CAAC,OAAO,CAAC,OAAO,CAAC,CAAC,CAAC;IACnD,OAAO,KAAK,CAAC,IAAI,EAAE,CAAC;AACxB,CAAC"}

package/packages/mailx-service/word-html.ts

@@ -0,0 +1,78 @@
+/**
+ * Word's own HTML export → a body fragment compose can take.
+ *
+ * 2026-09-13 20:10 EDT — Claude Code (Fable 5.1), at Bob's direction ("go
+ * ahead and build the export fragment route").
+ *
+ * The Edit-in-Word autosave sidecar calls Word's `Range.ExportFragment(path,
+ * wdFormatFilteredHTML)` over COM on every change, which writes Word's
+ * filtered-HTML rendering of the document WITHOUT changing the document's
+ * own file or format. That HTML keeps colours, fonts, alignment, spacing and
+ * paragraph borders as inline styles — the things mammoth's docx→HTML throws
+ * away (measured on a real reply 2026-09-13: mammoth returned only <p>, <a>
+ * and <br>). This module turns that export into a fragment: body content
+ * only, Word's class and comment cruft removed, sidecar image files inlined.
+ */
+
+import fs from "node:fs";
+import path from "node:path";
+
+/** Decode Word's export honouring its own charset meta (utf-8 in every
+ *  export seen so far, but Word has written windows-1252 in the past). */
+function decodeWordHtml(bytes: Buffer): string {
+    const probe = bytes.subarray(0, 2048).toString("latin1");
+    const m = /charset=["']?([\w-]+)/i.exec(probe);
+    const charset = (m?.[1] || "utf-8").toLowerCase();
+    try {
+        return new TextDecoder(charset).decode(bytes);
+    } catch (e: any) {
+        // An unknown label falls back to utf-8 rather than failing the
+        // mirror; the caller logs the fragment length either way.
+        console.warn(`  [word-edit] unknown charset "${charset}" in Word export — decoding as utf-8 (${e?.message || e})`);
+        return bytes.toString("utf-8");
+    }
+}
+
+const IMAGE_MIME: Record<string, string> = {
+    ".png": "image/png", ".jpg": "image/jpeg", ".jpeg": "image/jpeg", ".gif": "image/gif",
+    ".bmp": "image/bmp", ".webp": "image/webp", ".svg": "image/svg+xml", ".emf": "image/emf", ".wmf": "image/wmf",
+};
+
+/** Turn `<img src="x_files/image001.png">` into a data: URI. Word writes
+ *  images next to the export in a `<name>_files` folder; compose already
+ *  converts data: images to cid parts on send. A missing file is left as
+ *  is and logged — the text still mirrors. */
+function inlineImages(html: string, baseDir: string): string {
+    return html.replace(/<img\b[^>]*\bsrc\s*=\s*["']?([^"'\s>]+)["']?[^>]*>/gi, (tag, src: string) => {
+        if (/^(data|https?|cid):/i.test(src)) return tag;
+        const file = path.resolve(baseDir, decodeURIComponent(src));
+        const mime = IMAGE_MIME[path.extname(file).toLowerCase()];
+        if (!mime) return tag;
+        try {
+            const data = fs.readFileSync(file).toString("base64");
+            return tag.replace(src, `data:${mime};base64,${data}`);
+        } catch (e: any) {
+            console.warn(`  [word-edit] image ${src} from the Word export could not be read: ${e?.message || e}`);
+            return tag;
+        }
+    });
+}
+
+/** Read a Word filtered-HTML export and return body-content HTML for compose. */
+export function wordHtmlToFragment(htmPath: string): string {
+    const html = decodeWordHtml(fs.readFileSync(htmPath));
+    const bodyMatch = html.match(/<body[^>]*>([\s\S]*?)<\/body>/i);
+    let inner = bodyMatch ? bodyMatch[1] : html;
+    inner = inner
+        // Conditional comments around fake-list bullets, and any other comment.
+        .replace(/<!--[\s\S]*?-->/g, "")
+        // Word's paragraph/run classes; their styles live in the export's
+        // <style> block, which does not travel with the fragment.
+        .replace(/\s+class\s*=\s*["']?Mso[\w-]*["']?/gi, "")
+        // Office namespace tags (<o:p></o:p> paragraph marks) and their content-free shells.
+        .replace(/<\/?o:[^>]*>/gi, "")
+        // Word wraps the whole export in a section div; compose has no use for it.
+        .replace(/^\s*<div\s+class\s*=\s*["']?WordSection\d*["']?[^>]*>([\s\S]*)<\/div>\s*$/i, "$1");
+    inner = inlineImages(inner, path.dirname(htmPath));
+    return inner.trim();
+}