fix: preserve paired input evidence context (#204)

runners/agent-runner/src/output-protocol.ts

@@ -2,6 +2,7 @@ import fs from 'fs';
 import path from 'path';
 
 import {
+  expandImagePromptReferences,
   extractImageTagPaths,
   imageTagCaption,
   missingImageTagCaption,
@@ -79,7 +80,8 @@ export function buildMultimodalContent(
   text: string,
   log: LogFn,
 ): StreamContent {
-  const { imagePaths } = extractImageTagPaths(text);
+  const expandedText = expandImagePromptReferences(text);
+  const { imagePaths } = extractImageTagPaths(expandedText);
   if (imagePaths.length === 0) return text;
 
   const blocks: ContentBlock[] = [];
@@ -88,7 +90,7 @@ export function buildMultimodalContent(
     if (trimmed) blocks.push({ type: 'text', text: trimmed });
   };
 
-  for (const part of splitImageTagPromptParts(text)) {
+  for (const part of splitImageTagPromptParts(expandedText)) {
     if (part.type === 'text') {
       pushText(part.text);
       continue;

runners/agent-runner/test/output-protocol.test.ts

@@ -94,4 +94,57 @@ describe('agent-runner multimodal prompts', () => {
     ]);
     expect(logs).toContain(`Unsupported image type, skipping: ${imagePath}`);
   });
+
+  it('loads MEDIA image directives as Claude image blocks', () => {
+    const dir = fs.mkdtempSync(path.join(os.tmpdir(), 'ejclaw-media-image-'));
+    cleanupDirs.push(dir);
+    const imagePath = path.join(dir, 'media-render.png');
+    fs.writeFileSync(imagePath, ONE_PIXEL_PNG);
+
+    const content = buildMultimodalContent(`증거\nMEDIA:${imagePath}`, () => {
+      // no-op
+    });
+
+    expect(content).toEqual([
+      { type: 'text', text: '증거' },
+      { type: 'text', text: 'Image evidence: media-render.png' },
+      {
+        type: 'image',
+        source: {
+          type: 'base64',
+          media_type: 'image/png',
+          data: ONE_PIXEL_PNG.toString('base64'),
+        },
+      },
+    ]);
+  });
+
+  it('loads markdown image links as Claude image blocks', () => {
+    const dir = fs.mkdtempSync(
+      path.join(os.tmpdir(), 'ejclaw-markdown-image-'),
+    );
+    cleanupDirs.push(dir);
+    const imagePath = path.join(dir, 'markdown-render.png');
+    fs.writeFileSync(imagePath, ONE_PIXEL_PNG);
+
+    const content = buildMultimodalContent(
+      `증거 ![render](${imagePath})`,
+      () => {
+        // no-op
+      },
+    );
+
+    expect(content).toEqual([
+      { type: 'text', text: '증거' },
+      { type: 'text', text: 'Image evidence: markdown-render.png' },
+      {
+        type: 'image',
+        source: {
+          type: 'base64',
+          media_type: 'image/png',
+          data: ONE_PIXEL_PNG.toString('base64'),
+        },
+      },
+    ]);
+  });
 });

runners/codex-runner/src/app-server-input.ts

@@ -2,6 +2,7 @@ import fs from 'fs';
 import path from 'path';
 
 import {
+  expandImagePromptReferences,
   extractImageTagPaths,
   imageTagCaption,
   missingImageTagCaption,
@@ -22,7 +23,8 @@ export function parseAppServerInput(
   text: string,
   log: (message: string) => void = () => undefined,
 ): AppServerInputItem[] {
-  const { imagePaths } = extractImageTagPaths(text);
+  const expandedText = expandImagePromptReferences(text);
+  const { imagePaths } = extractImageTagPaths(expandedText);
   const input: AppServerInputItem[] = [];
   const pushText = (value: string) => {
     const trimmed = value.trim();
@@ -30,7 +32,7 @@ export function parseAppServerInput(
   };
 
   if (imagePaths.length > 0) {
-    for (const part of splitImageTagPromptParts(text)) {
+    for (const part of splitImageTagPromptParts(expandedText)) {
       if (part.type === 'text') {
         pushText(part.text);
         continue;

runners/codex-runner/test/app-server-input.test.ts

@@ -86,4 +86,36 @@ describe('codex app-server input', () => {
     ]);
     expect(logs).toContain(`Unsupported image type, skipping: ${imagePath}`);
   });
+
+  it('loads MEDIA image directives as local image input items', () => {
+    const dir = fs.mkdtempSync(path.join(os.tmpdir(), 'ejclaw-codex-media-'));
+    cleanupDirs.push(dir);
+    const imagePath = path.join(dir, 'media-render.png');
+    fs.writeFileSync(imagePath, ONE_PIXEL_PNG);
+
+    const input = parseAppServerInput(`증거\nMEDIA:${imagePath}`);
+
+    expect(input).toEqual([
+      { type: 'text', text: '증거' },
+      { type: 'text', text: 'Image evidence: media-render.png' },
+      { type: 'localImage', path: imagePath },
+    ]);
+  });
+
+  it('loads markdown image links as local image input items', () => {
+    const dir = fs.mkdtempSync(
+      path.join(os.tmpdir(), 'ejclaw-codex-markdown-'),
+    );
+    cleanupDirs.push(dir);
+    const imagePath = path.join(dir, 'markdown-render.png');
+    fs.writeFileSync(imagePath, ONE_PIXEL_PNG);
+
+    const input = parseAppServerInput(`증거 ![render](${imagePath})`);
+
+    expect(input).toEqual([
+      { type: 'text', text: '증거' },
+      { type: 'text', text: 'Image evidence: markdown-render.png' },
+      { type: 'localImage', path: imagePath },
+    ]);
+  });
 });

runners/shared/src/agent-protocol.ts

@@ -92,6 +92,36 @@ export function extractImageTagPaths(text: string): {
   };
 }
 
+export function expandImagePromptReferences(text: string): string {
+  const codeSpans = fencedCodeSpans(text);
+  const withMediaImages = text.replace(
+    MEDIA_TAG_RE,
+    (full: string, doubleQuoted, singleQuoted, backticked, bare, offset) => {
+      if (isInsideSpans(offset, codeSpans)) return full;
+      const filePath = String(
+        doubleQuoted ?? singleQuoted ?? backticked ?? bare ?? '',
+      ).trim();
+      if (!filePath.startsWith('/') || !IMAGE_EXT_RE.test(filePath)) {
+        return full;
+      }
+      const name = attachmentName(filePath) ?? filePath;
+      return `[Image: ${name} → ${filePath}]`;
+    },
+  );
+
+  const markdownCodeSpans = fencedCodeSpans(withMediaImages);
+  return withMediaImages.replace(
+    MARKDOWN_IMAGE_ABSOLUTE_LINK_RE,
+    (full: string, rawPath: string, offset: number) => {
+      if (isInsideSpans(offset, markdownCodeSpans)) return full;
+      const filePath = rawPath.trim();
+      if (!IMAGE_EXT_RE.test(filePath)) return full;
+      const name = attachmentName(filePath) ?? filePath;
+      return `[Image: ${name} → ${filePath}]`;
+    },
+  );
+}
+
 export type ImageTagPromptPart =
   | { type: 'text'; text: string }
   | { type: 'image'; label: string | null; path: string; raw: string };

runners/shared/src/index.ts

@@ -38,6 +38,7 @@ export {
   type TaskContextMode,
 } from './task-runtime.js';
 export {
+  expandImagePromptReferences,
   extractMarkdownImageAttachments,
   extractMediaAttachments,
   extractImageTagPaths,

runners/shared/test/agent-protocol.test.ts

@@ -1,6 +1,7 @@
 import { describe, expect, it } from 'vitest';
 
 import {
+  expandImagePromptReferences,
   extractImageTagPaths,
   imageTagCaption,
   missingImageTagCaption,
@@ -59,6 +60,23 @@ describe('shared agent protocol helpers', () => {
     );
   });
 
+  it('expands MEDIA image directives into image prompt tags', () => {
+    expect(expandImagePromptReferences('증거\nMEDIA:/tmp/render.png\n끝')).toBe(
+      '증거\n[Image: render.png → /tmp/render.png]\n끝',
+    );
+  });
+
+  it('expands markdown image links into image prompt tags', () => {
+    expect(
+      expandImagePromptReferences('증거 ![render](/tmp/render.png) 끝'),
+    ).toBe('증거 [Image: render.png → /tmp/render.png] 끝');
+  });
+
+  it('keeps non-image media and fenced media text unchanged', () => {
+    const text = 'MEDIA:/tmp/demo.mp4\n```text\nMEDIA:/tmp/render.png\n```';
+    expect(expandImagePromptReferences(text)).toBe(text);
+  });
+
   it('normalizes plain text runner output as public text', () => {
     expect(normalizePublicTextOutput('DONE')).toEqual({
       result: 'DONE',