@bacnh85/pi-web 0.6.1 → 0.6.2

package/CHANGELOG.md

@@ -1,6 +1,22 @@
 # Changelog
 
-## 0.6.2 (2026-08-30)
+## 0.6.2 (2026-09-06)
+
+### Changed
+
+- `web_screenshot` now returns the PNG **inline as an image block**
+  (`ImageContent`) alongside the text summary, so multimodal models (GLM-5.3,
+  Claude, Gemini) actually see the screenshot instead of a base64 char
+  count. The "Data: base64 PNG (N chars)" line is gone; artifact/MIME/size
+  summary unchanged. Inspired by zcode-plugins video2code's vision-in-the-loop.
+  Regression-tested in `test/unit/screenshot.test.ts` (fetch stubbed — no
+  daemon needed): image block present + base64-text line absent; text-only
+  fallback when the daemon returns no screenshot.
+
+  Daemon `success:false` responses (HTTP 200) now surface `error_message` as
+  a tool error instead of returning a silently empty screenshot result.
+
+## 0.6.1 (2026-08-30)
 
 ### Changed
 
@@ -11,9 +27,7 @@
   hook.test.ts assertion updated to the compressed phrasing. No tool,
   parameter, or default changed.
 
-## 0.6.1 (2026-08-19)
-
-### Changed
+### Changed (2026-08-19)
 
 - `web_extract` agy backend default model updated to `gemini-3.7-flash-medium`
   — the current Flash generation in agy 1.1.x (3.6 is still served, this just

package/README.md

@@ -142,7 +142,7 @@ web_crawl url="https://example.com" mode=light poll=true    # Poll for completio
 
 ### `web_screenshot` — Page screenshot
 
-Captures a full-page PNG screenshot using Crawl4AI. Returns base64-encoded PNG.
+Captures a full-page PNG screenshot using Crawl4AI. Returns the PNG inline as an image block (multimodal models see it); text summary includes artifact/MIME/size.
 
 ```
 web_screenshot url="https://example.com"

package/extensions/index.ts

@@ -286,9 +286,9 @@ export default function piWebExtension(pi: ExtensionAPI) {
     name: "web_screenshot",
     label: "Web Page Screenshot",
     description:
-      "Full-page PNG screenshot via Crawl4AI.",
+      "Full-page PNG screenshot via Crawl4AI. The PNG is returned inline as an image block.",
     promptSnippet: "Screenshot a webpage",
-    promptGuidelines: ["Full-page PNG; use when web_extract fails on JS-heavy pages."],
+    promptGuidelines: ["Full-page PNG returned inline (multimodal models see it); use when web_extract fails on JS-heavy pages, or to visually inspect a built UI."],
     parameters: Type.Object({
       url: Type.String(),
       wait_for: Type.Optional(Type.Number({ default: 2, description: "Seconds to wait before capture." })),
@@ -305,16 +305,23 @@ export default function piWebExtension(pi: ExtensionAPI) {
         params.wait_for_images as boolean | undefined,
         signal,
       );
+      if (result.success === false) {
+        throw new Error(String(result.error_message ?? "Crawl4AI screenshot failed"));
+      }
       const screenshot = result.screenshot as string | undefined;
       const artifactUrl = result.url as string | undefined;
       const mime = result.mime as string | undefined;
       const size = result.size as number | undefined;
       let text = `Screenshot: ${params.url}\n`;
-      if (screenshot) text += `Data: base64 PNG (${screenshot.length} chars)\n`;
       if (artifactUrl) text += `Artifact: ${artifactUrl}\n`;
       if (mime) text += `MIME: ${mime}\n`;
       if (size) text += `Size: ${size} bytes\n`;
-      return { content: [{ type: "text" as const, text: truncateText(text) }], details: { ...result, url: params.url } };
+      // Return the PNG as a real image block so multimodal models see it.
+      const content: Array<{ type: "text"; text: string } | { type: "image"; data: string; mimeType: string }> = [
+        { type: "text", text: truncateText(text) },
+      ];
+      if (screenshot) content.push({ type: "image", data: screenshot, mimeType: mime || "image/png" });
+      return { content, details: { ...result, url: params.url } };
     },
   });
 

package/extensions/lib/content.ts

@@ -59,6 +59,16 @@ export async function fetchReadableContent(
     signal: signalWithTimeout(timeoutMs, signal),
   });
   if (!response.ok) throw new Error(`HTTP ${response.status}: ${response.statusText}`);
+  // Raw text/JSON payloads (raw.githubusercontent.com, JSON APIs) — Readability
+  // shreds them to nothing. Pass through verbatim. text/html and text/xml keep
+  // the Readability path — they're ordinary web pages. Session mining: 9/40
+  // static extract failures were raw-text/JSON shapes.
+  const contentType = (response.headers.get("content-type") ?? "").split(";")[0].trim();
+  if ((contentType.startsWith("text/") && contentType !== "text/html" && contentType !== "text/xml") || contentType === "application/json") {
+    const body = await response.text();
+    const markdown = contentType === "application/json" ? "```json\n" + body + "\n```" : body;
+    return { title: "", markdown: markdown.slice(0, 20000) };
+  }
   const html = await response.text();
   const deps = loadReadableContentDependencies();
   const dom = new deps.JSDOM(html, { url });

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@bacnh85/pi-web",
-  "version": "0.6.1",
+  "version": "0.6.2",
   "description": "Pi extension for web search, page extraction, Firecrawl scraping/crawling, and Crawl4AI headless browser crawling.",
   "type": "module",
   "license": "MIT",