fix(reply): parse markdown image replies as media

vincentkoc · vincentkoc · commit 2dfcd77e71b8 · 2026-04-14T10:50:16.000+01:00
diff --git a/CHANGELOG.md b/CHANGELOG.md
@@ -26,6 +26,7 @@ Docs: https://docs.openclaw.ai
 - Agents/subagents: emit the subagent registry lazy-runtime stub on the stable dist path that both source and bundled runtime imports resolve, so the follow-up dist fix no longer still fails with `ERR_MODULE_NOT_FOUND` at runtime. (#66420) Thanks @obviyus.
 - Media-understanding/proxy env: auto-upgrade provider HTTP helper requests to trusted env-proxy mode only when `HTTP_PROXY`/`HTTPS_PROXY` is active and the target is not bypassed by `NO_PROXY`, so remote media-understanding and transcription requests stop failing local DNS pre-resolution in proxy-only environments without widening SSRF bypasses. (#52162) Thanks @mjamiv and @vincentkoc.
 - Telegram/media downloads: let Telegram media fetches trust an operator-configured explicit proxy for target DNS resolution after hostname-policy checks, so proxy-backed installs stop failing `could not download media` on Bot API file downloads after the DNS-pinning regression. (#66245) Thanks @dawei41468 and @vincentkoc.
+- Telegram/media replies: parse markdown image syntax into outbound media payloads on the final reply path, so Telegram group chats stop falling back to plain-text image URLs when the model or a tool emits `![...](...)` instead of a `MEDIA:` token. (#66191) Thanks @apezam and @vincentkoc.
 - Browser: keep loopback CDP readiness checks reachable under strict SSRF defaults so OpenClaw can reconnect to locally started managed Chrome. (#66354) Thanks @hxy91819.
 - Agents/context engine: compact engine-owned sessions from the first tool-loop delta and preserve ingest fallback when `afterTurn` is absent, so long-running tool loops can stay bounded without dropping engine state. (#63555) Thanks @Bikkies.
 - Discord/native commands: return the real status card for native `/status` interactions instead of falling through to the synthetic `✅ Done.` ack when the generic dispatcher produces no visible reply. (#54629) Thanks @tkozzer and @vincentkoc.
diff --git a/src/auto-reply/reply/agent-runner-payloads.test.ts b/src/auto-reply/reply/agent-runner-payloads.test.ts
@@ -220,6 +220,20 @@ describe("buildReplyPayloads media filter integration", () => {
     expect(replyPayloads).toHaveLength(0);
   });
 
+  it("extracts markdown image replies into final payload media urls", async () => {
+    const { replyPayloads } = await buildReplyPayloads({
+      ...baseParams,
+      payloads: [{ text: "Here you go\n\n![chart](https://example.com/chart.png)" }],
+    });
+
+    expect(replyPayloads).toHaveLength(1);
+    expect(replyPayloads[0]).toMatchObject({
+      text: "Here you go",
+      mediaUrl: "https://example.com/chart.png",
+      mediaUrls: ["https://example.com/chart.png"],
+    });
+  });
+
   it("deduplicates final payloads against directly sent block keys regardless of replyToId", async () => {
     // When block streaming is not active but directlySentBlockKeys has entries
     // (e.g. from pre-tool flush), the key should match even if replyToId differs.
diff --git a/src/media/parse.test.ts b/src/media/parse.test.ts
@@ -103,4 +103,21 @@ describe("splitMediaFromOutput", () => {
       { type: "text", text: "```text\nMEDIA:https://example.com/ignored.png\n```\nAfter" },
     ]);
   });
+
+  it("extracts markdown image urls while keeping surrounding caption text", () => {
+    expectParsedMediaOutputCase("Caption\n\n![chart](https://example.com/chart.png)", {
+      text: "Caption",
+      mediaUrls: ["https://example.com/chart.png"],
+    });
+  });
+
+  it("extracts multiple markdown image urls in order", () => {
+    expectParsedMediaOutputCase(
+      "Before\n![one](https://example.com/one.png)\nMiddle\n![two](./out/two.png)\nAfter",
+      {
+        text: "Before\nMiddle\nAfter",
+        mediaUrls: ["https://example.com/one.png", "./out/two.png"],
+      },
+    );
+  });
 });
diff --git a/src/media/parse.ts b/src/media/parse.ts
@@ -5,6 +5,7 @@ import { parseAudioTag } from "./audio-tags.js";
 
 // Allow optional wrapping backticks and punctuation after the token; capture the core token.
 export const MEDIA_TOKEN_RE = /\bMEDIA:\s*`?([^\n]+)`?/gi;
+const MARKDOWN_IMAGE_RE = /!\[[^\]]*]\(([^)\n]+)\)/g;
 
 export type ParsedMediaOutputSegment =
   | {
@@ -125,6 +126,59 @@ function mayContainFenceMarkers(input: string): boolean {
   return input.includes("```") || input.includes("~~~");
 }
 
+function cleanLineText(text: string): string {
+  return text.replace(/[ \t]{2,}/g, " ").trim();
+}
+
+function collectMarkdownImageSegments(params: { line: string; media: string[] }): {
+  cleanedLine?: string;
+  lineSegments: ParsedMediaOutputSegment[];
+  foundMedia: boolean;
+} {
+  const matches = Array.from(params.line.matchAll(MARKDOWN_IMAGE_RE));
+  if (matches.length === 0) {
+    return { lineSegments: [], foundMedia: false };
+  }
+
+  const pieces: string[] = [];
+  const lineSegments: ParsedMediaOutputSegment[] = [];
+  let cursor = 0;
+  let foundMedia = false;
+
+  for (const match of matches) {
+    const start = match.index ?? 0;
+    pieces.push(params.line.slice(cursor, start));
+
+    const target = normalizeMediaSource(cleanCandidate(unwrapQuoted(match[1]) ?? match[1] ?? ""));
+    if (isValidMedia(target, { allowSpaces: true, allowBareFilename: true })) {
+      const beforeText = cleanLineText(pieces.join(""));
+      if (beforeText) {
+        lineSegments.push({ type: "text", text: beforeText });
+      }
+      pieces.length = 0;
+      params.media.push(target);
+      lineSegments.push({ type: "media", url: target });
+      foundMedia = true;
+    } else {
+      pieces.push(match[0]);
+    }
+
+    cursor = start + match[0].length;
+  }
+
+  pieces.push(params.line.slice(cursor));
+  const cleanedLine = cleanLineText(pieces.join(""));
+  if (cleanedLine) {
+    lineSegments.push({ type: "text", text: cleanedLine });
+  }
+
+  return {
+    cleanedLine: cleanedLine || undefined,
+    lineSegments,
+    foundMedia,
+  };
+}
+
 // Check if a character offset is inside any fenced code block
 function isInsideFence(fenceSpans: Array<{ start: number; end: number }>, offset: number): boolean {
   return fenceSpans.some((span) => offset >= span.start && offset < span.end);
@@ -144,8 +198,9 @@ export function splitMediaFromOutput(raw: string): {
     return { text: "" };
   }
   const mayContainMediaToken = /media:/i.test(trimmedRaw);
+  const mayContainMarkdownImage = /!\[[^\]]*]\(/.test(trimmedRaw);
   const mayContainAudioTag = trimmedRaw.includes("[[");
-  if (!mayContainMediaToken && !mayContainAudioTag) {
+  if (!mayContainMediaToken && !mayContainMarkdownImage && !mayContainAudioTag) {
     return { text: trimmedRaw };
   }
 
@@ -185,8 +240,23 @@ export function splitMediaFromOutput(raw: string): {
 
     const trimmedStart = line.trimStart();
     if (!trimmedStart.startsWith("MEDIA:")) {
-      keptLines.push(line);
-      pushTextSegment(line);
+      const markdownImageResult = collectMarkdownImageSegments({ line, media });
+      if (!markdownImageResult.foundMedia) {
+        keptLines.push(line);
+        pushTextSegment(line);
+      } else {
+        foundMediaToken = true;
+        if (markdownImageResult.cleanedLine) {
+          keptLines.push(markdownImageResult.cleanedLine);
+        }
+        for (const segment of markdownImageResult.lineSegments) {
+          if (segment.type === "text") {
+            pushTextSegment(segment.text);
+            continue;
+          }
+          segments.push(segment);
+        }
+      }
       lineOffset += line.length + 1; // +1 for newline
       continue;
     }
@@ -269,10 +339,7 @@ export function splitMediaFromOutput(raw: string): {
       }
 
       if (hasValidMedia) {
-        const beforeText = pieces
-          .join("")
-          .replace(/[ \t]{2,}/g, " ")
-          .trim();
+        const beforeText = cleanLineText(pieces.join(""));
         if (beforeText) {
           lineSegments.push({ type: "text", text: beforeText });
         }
@@ -297,10 +364,7 @@ export function splitMediaFromOutput(raw: string): {
 
     pieces.push(line.slice(cursor));
 
-    const cleanedLine = pieces
-      .join("")
-      .replace(/[ \t]{2,}/g, " ")
-      .trim();
+    const cleanedLine = cleanLineText(pieces.join(""));
 
     // If the line becomes empty, drop it.
     if (cleanedLine) {
