@mkterswingman/5mghost-wonder 0.0.15 → 0.0.17

package/dist/wecom/browser-probe.js

@@ -10,6 +10,8 @@ const CDP_CALL_TIMEOUT_MS = 5000;
 const BROWSER_START_TIMEOUT_MS = 20_000;
 const DEFAULT_CAPTURE_MS = 8_000;
 const MAX_NETWORK_CANDIDATES = 40;
+const MAX_IMAGE_DOWNLOADS = 20;
+const MAX_IMAGE_BYTES = 25 * 1024 * 1024;
 let cdpNextId = 1;
 export async function runBrowserNoExportProbe(options) {
     const timeoutMs = options.timeoutMs ?? DEFAULT_CAPTURE_MS;
@@ -157,6 +159,18 @@ export async function runBrowserNoExportRead(options) {
                 evidence.push("opendoc-response");
             if (extracted?.text)
                 evidence.push("initial-attributed-text");
+            if (extracted?.structureHints)
+                evidence.push("structure-hints");
+            const images = options.saveDir && extracted?.imageUrls
+                ? await downloadImageResources(extracted.imageUrls, options.saveDir)
+                : (extracted?.imageUrls ?? []).map((url) => ({
+                    url,
+                    ...parseImageDimensionsFromUrl(url),
+                    status: "skipped",
+                    error: "pass --save <dir> to download image resources",
+                }));
+            if (images.some((image) => image.status === "downloaded"))
+                evidence.push("image-resource-download");
             const result = {
                 mode: "browser-no-export-read",
                 status: extracted?.text ? "partial" : "fail",
@@ -167,6 +181,8 @@ export async function runBrowserNoExportRead(options) {
                 text: extracted?.text,
                 textLength: extracted?.text.length,
                 imageUrls: extracted?.imageUrls,
+                images,
+                structureHints: extracted?.structureHints,
                 extraction: extracted?.text ? "opendoc-initial-attributed-text" : "none",
                 evidence,
                 warnings: extracted?.warnings ?? [
@@ -174,7 +190,7 @@ export async function runBrowserNoExportRead(options) {
                 ],
                 missing: [
                     "table merge ranges",
-                    ...(extracted?.imageUrls && extracted.imageUrls.length > 0 ? [] : ["image original resources"]),
+                    ...(images.some((image) => image.status === "downloaded") ? [] : ["image original resources"]),
                     "image anchors",
                     "floating vs fixed image classification",
                 ],
@@ -292,16 +308,24 @@ export function extractTextFromOpendoc(body) {
     const decodedChunks = chunks
         .filter((chunk) => typeof chunk === "string")
         .flatMap((chunk) => extractInitialAttributedTextCandidates(chunk));
+    const decodedText = decodedChunks.join("\n");
     const textCandidate = chooseTextCandidate(decodedChunks);
-    const text = normalizeExtractedText(textCandidate ?? decodedChunks.join("\n"));
+    const text = normalizeExtractedText(textCandidate ?? decodedText);
     if (!text)
         return null;
-    const imageUrls = extractImageUrls(decodedChunks.join("\n"));
+    const imageUrls = extractImageUrls(decodedText);
+    const structureHints = analyzeOpendocStructure({
+        opendocBody: body,
+        htmlData: typeof payload.htmlData === "string" ? payload.htmlData : "",
+        decodedText,
+        decodedCandidateCount: decodedChunks.length,
+    });
     return {
         title: payload.clientVars?.title ?? payload.clientVars?.initialTitle,
         padType: payload.clientVars?.padType ?? payload.padType,
         text,
         imageUrls,
+        structureHints,
         warnings: [
             "No-export text is decoded from opendoc initialAttributedText; rich styles are not reconstructed yet.",
             ...(imageUrls.length > 0
@@ -310,6 +334,27 @@ export function extractTextFromOpendoc(body) {
         ],
     };
 }
+function analyzeOpendocStructure(input) {
+    return {
+        opendoc: summarizeStructureSignals(input.opendocBody),
+        htmlData: {
+            ...summarizeStructureSignals(input.htmlData),
+            hasTableMarkup: /<\/?(table|tr|td|th)\b/i.test(input.htmlData),
+        },
+        initialAttributedText: {
+            ...summarizeStructureSignals(input.decodedText),
+            decodedCandidateCount: input.decodedCandidateCount,
+        },
+    };
+}
+function summarizeStructureSignals(value) {
+    return {
+        length: value.length,
+        hasTableSignals: /\b(table|cell|row|column|sheet|workbook)\b/i.test(value),
+        hasMergeSignals: /\b(merge|merged|rowspan|colspan|mergeCell|mergeCells)\b/i.test(value),
+        hasImageSignals: /\b(image|img|pic|picture|media|download_url|qpic)\b/i.test(value),
+    };
+}
 function extractInitialAttributedTextCandidates(chunk) {
     let decoded;
     try {
@@ -426,6 +471,97 @@ function extractImageUrls(text) {
         .map((url) => url.replace(/[)*,.;:]+$/g, ""))
         .filter((url) => /qpic\.cn|weixin\.qq\.com|doc\.weixin\.qq\.com/i.test(url))));
 }
+async function downloadImageResources(imageUrls, saveDir) {
+    const imageDir = resolve(saveDir, "images");
+    mkdirSync(imageDir, { recursive: true });
+    const results = [];
+    for (const [index, url] of imageUrls.slice(0, MAX_IMAGE_DOWNLOADS).entries()) {
+        const dimensions = parseImageDimensionsFromUrl(url);
+        try {
+            const res = await fetch(url);
+            if (!res.ok) {
+                results.push({
+                    url,
+                    ...dimensions,
+                    status: "failed",
+                    error: `HTTP ${res.status}`,
+                });
+                continue;
+            }
+            const contentType = res.headers.get("content-type") ?? undefined;
+            const bytes = Buffer.from(await res.arrayBuffer());
+            if (bytes.length > MAX_IMAGE_BYTES) {
+                results.push({
+                    url,
+                    ...dimensions,
+                    contentType,
+                    sizeBytes: bytes.length,
+                    status: "failed",
+                    error: `image exceeds ${MAX_IMAGE_BYTES} bytes`,
+                });
+                continue;
+            }
+            const filePath = resolve(imageDir, `image-${String(index + 1).padStart(3, "0")}${extensionForContentType(contentType)}`);
+            writeFileSync(filePath, bytes, { mode: 0o600 });
+            results.push({
+                url,
+                path: filePath,
+                contentType,
+                sizeBytes: bytes.length,
+                ...dimensions,
+                status: "downloaded",
+            });
+        }
+        catch (err) {
+            results.push({
+                url,
+                ...dimensions,
+                status: "failed",
+                error: err instanceof Error ? err.message : String(err),
+            });
+        }
+    }
+    if (imageUrls.length > MAX_IMAGE_DOWNLOADS) {
+        for (const url of imageUrls.slice(MAX_IMAGE_DOWNLOADS)) {
+            results.push({
+                url,
+                ...parseImageDimensionsFromUrl(url),
+                status: "skipped",
+                error: `only first ${MAX_IMAGE_DOWNLOADS} images are downloaded`,
+            });
+        }
+    }
+    return results;
+}
+export function parseImageDimensionsFromUrl(url) {
+    try {
+        const parsed = new URL(url);
+        const width = Number(parsed.searchParams.get("w"));
+        const height = Number(parsed.searchParams.get("h"));
+        return {
+            width: Number.isFinite(width) && width > 0 ? width : undefined,
+            height: Number.isFinite(height) && height > 0 ? height : undefined,
+        };
+    }
+    catch {
+        return {};
+    }
+}
+function extensionForContentType(contentType) {
+    if (!contentType)
+        return ".bin";
+    if (contentType.includes("png"))
+        return ".png";
+    if (contentType.includes("jpeg") || contentType.includes("jpg"))
+        return ".jpg";
+    if (contentType.includes("gif"))
+        return ".gif";
+    if (contentType.includes("webp"))
+        return ".webp";
+    if (contentType.includes("svg"))
+        return ".svg";
+    return ".bin";
+}
 function normalizeExtractedText(text) {
     return text
         .replace(/p\.\d{8,}/g, "")

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@mkterswingman/5mghost-wonder",
-  "version": "0.0.15",
+  "version": "0.0.17",
   "description": "企微文档读取 CLI — WeCom document reader",
   "type": "module",
   "engines": {

package/skills/setup-5mghost-wonder/SKILL.md

@@ -7,9 +7,9 @@ description: Use this skill when the user wants to install or set up wonder, say
 
 ## Skill version
 
-This skill matches **wonder 0.0.15**.
+This skill matches **wonder 0.0.17**.
 
-Once the CLI is installed in Step 1, run `wonder --version`. If the output does not equal `0.0.15`, the CLI on disk has drifted from the skill text loaded in this session. Ask the user to run `/update-5mghost-wonder`, then **start a fresh AI session** (`/exit` and re-enter, or open a new chat) — skill text already loaded into a running session does not refresh after `wonder update`, even though the file on disk has been replaced.
+Once the CLI is installed in Step 1, run `wonder --version`. If the output does not equal `0.0.17`, the CLI on disk has drifted from the skill text loaded in this session. Ask the user to run `/update-5mghost-wonder`, then **start a fresh AI session** (`/exit` and re-enter, or open a new chat) — skill text already loaded into a running session does not refresh after `wonder update`, even though the file on disk has been replaced.
 
 After a successful first install, also remind the user to start a fresh AI session before invoking `/use-5mghost-wonder` for the first time. The skill files were just written to disk; the current session never loaded them.
 

package/skills/update-5mghost-wonder/SKILL.md

@@ -7,7 +7,7 @@ description: Use this skill when the user wants to update or upgrade wonder, say
 
 ## Skill version
 
-This skill matches **wonder 0.0.15**.
+This skill matches **wonder 0.0.17**.
 
 ---
 

package/skills/use-5mghost-wonder/SKILL.md

@@ -10,10 +10,10 @@ the referenced workflow files needed for the current task.
 
 ## Version Gate
 
-This skill matches **wonder 0.0.15**.
+This skill matches **wonder 0.0.17**.
 
 On first use in a session, follow `references/session-init.md`. If the installed
-CLI version differs from `0.0.15`, stop and ask the user to run
+CLI version differs from `0.0.17`, stop and ask the user to run
 `/update-5mghost-wonder`, then start a fresh AI session.
 
 ## Hard Rules

package/skills/use-5mghost-wonder/references/read-no-export-browser.md

@@ -37,8 +37,10 @@ wonder browser read <url> --save /tmp/wonder-browser-read
 ```
 
 This currently attempts text extraction from the browser `opendoc`
-`initialAttributedText`. Treat it as `partial` unless it also proves image and
-table structure.
+`initialAttributedText`, downloads detected image resources when `--save` is
+present, and returns safe `structureHints` for `htmlData`,
+`initialAttributedText`, and the full `opendoc` body. Treat it as `partial`
+unless it also proves actual table cells, merge ranges, and image anchors.
 
 3. If the read is insufficient, run the evidence probe:
 
@@ -59,6 +61,8 @@ Use `--headed` only when login/debug visibility is needed.
    - WebSocket messages if they expose structured document operations
    - screenshots only as a fallback or visual cross-check
 6. For tables, specifically look for:
+   - `structureHints.*.hasTableSignals` and `hasMergeSignals`
+   - `structureHints.htmlData.hasTableMarkup`
    - cell coordinates
    - displayed text
    - merge ranges
@@ -82,6 +86,12 @@ Return a concise result:
   "mergeRangesAvailable": true,
   "imagesAvailable": true,
   "imageAnchorsAvailable": true,
+  "structureHints": {
+    "htmlData": {
+      "hasTableMarkup": true,
+      "hasMergeSignals": true
+    }
+  },
   "evidence": ["runtime-store", "network-response", "websocket", "screenshot"],
   "missing": []
 }