fix: forward image attachments for OpenAI-compatible providers (#373)

## Summary
- Treat OpenAI-compatible chat, Anthropic, OpenAI Responses, and Codex
wires as image-capable when synthesizing provider models.
- Forward image attachments through the v0.2 agent path for
OpenAI-compatible providers instead of relying on model-id heuristics.
- Add focused core/providers tests and a changeset.

Fixes #201

## Verification
- pnpm --filter @open-codesign/core test -- src/agent.test.ts
- pnpm --filter @open-codesign/providers test -- src/index.test.ts
- pnpm --filter @open-codesign/core typecheck
- pnpm --filter @open-codesign/providers typecheck

Co-authored-by: Sun-sunshine06 <Sun-sunshine06@users.noreply.github.com>

Author: Sun-sunshine06

.changeset/openai-chat-image-attachments.md

@@ -0,0 +1,6 @@
+---
+"@open-codesign/core": patch
+"@open-codesign/providers": patch
+---
+
+Forward image attachments through OpenAI-compatible and Anthropic-style provider paths instead of only marking Codex synthesized models as image-capable.

apps/desktop/src/renderer/src/components/AddCustomProviderModal.tsx

@@ -298,7 +298,7 @@ export function AddCustomProviderModal({
         if (!editTarget.builtin) {
           const previous = editTarget.tlsRejectUnauthorized === true;
           if (previous !== tlsRejectUnauthorized) {
-            update.tlsRejectUnauthorized = tlsRejectUnauthorized ? true : false;
+            update.tlsRejectUnauthorized = !!tlsRejectUnauthorized;
           }
         }
         await window.codesign.config.updateProvider(update);

biome.json

@@ -1,5 +1,5 @@
 {
-  "$schema": "https://biomejs.dev/schemas/2.4.14/schema.json",
+  "$schema": "https://biomejs.dev/schemas/2.4.15/schema.json",
   "vcs": {
     "enabled": true,
     "clientKind": "git",

packages/core/src/agent.test.ts

@@ -1156,6 +1156,33 @@ describe('generateViaAgent()', () => {
     );
   });
 
+  it('passes image attachments through openai-compatible agent models', async () => {
+    scriptedAgent = { assistantText: RESPONSE_WITH_ARTIFACT };
+    await generateViaAgent(
+      {
+        prompt: 'replicate this screenshot',
+        history: [],
+        model: { provider: 'custom-openai', modelId: 'local-text-or-vision-model' },
+        apiKey: 'sk-test',
+        wire: 'openai-chat',
+        baseUrl: 'https://gateway.example.test/v1',
+        attachments: [
+          {
+            name: 'shot.png',
+            path: 'references/shot.png',
+            mediaType: 'image/png',
+            imageDataUrl: 'data:image/png;base64,aW1n',
+          },
+        ],
+      },
+      { fs: makeStubFs({}) },
+    );
+
+    expect(agentCalls[0]?.prompts[0]?.images).toEqual([
+      { type: 'image', data: 'aW1n', mimeType: 'image/png' },
+    ]);
+  });
+
   it('blocks preview and done until set_todos has run for fresh multi-step work', async () => {
     scriptedAgent = { assistantText: RESPONSE_WITH_ARTIFACT };
     await generateViaAgent(

packages/core/src/agent.ts

@@ -242,6 +242,9 @@ function supportsImageInput(wire: WireApi | undefined, modelId: string): boolean
   if (wire === 'anthropic' || wire === 'openai-responses' || wire === 'openai-codex-responses') {
     return true;
   }
+  if (wire === 'openai-chat') {
+    return true;
+  }
   const lower = modelId.toLowerCase();
   return (
     lower.includes('vision') ||

packages/providers/src/index.test.ts

@@ -361,6 +361,57 @@ describe('complete', () => {
     expect(result.content).toBe('ok');
   });
 
+  it('keeps image inputs for synthesized openai-chat models', async () => {
+    getModelMock.mockReturnValue(undefined);
+    completeSimpleMock.mockImplementationOnce(async (model, context) => {
+      expect(model).toMatchObject({
+        api: 'openai-completions',
+        input: ['text', 'image'],
+        baseUrl: 'https://gateway.example.test/v1',
+      });
+      expect(context.messages).toEqual([
+        {
+          role: 'user',
+          content: [
+            { type: 'text', text: 'use this screenshot' },
+            { type: 'image', data: 'AAAA', mimeType: 'image/png' },
+          ],
+          timestamp: 1,
+        },
+      ]);
+      return {
+        role: 'assistant',
+        content: [{ type: 'text', text: 'ok' }],
+        api: 'openai-completions',
+        provider: 'custom-openai',
+        model: 'local-text-or-vision-model',
+        usage: {
+          input: 1,
+          output: 1,
+          cacheRead: 0,
+          cacheWrite: 0,
+          totalTokens: 2,
+          cost: { input: 0, output: 0, cacheRead: 0, cacheWrite: 0, total: 0 },
+        },
+        stopReason: 'stop',
+        timestamp: Date.now(),
+      };
+    });
+
+    const result = await complete(
+      { provider: 'custom-openai', modelId: 'local-text-or-vision-model' },
+      [{ role: 'user', content: 'use this screenshot' }],
+      {
+        apiKey: 'sk-test',
+        wire: 'openai-chat',
+        baseUrl: 'https://gateway.example.test/v1',
+        userImages: [{ data: 'AAAA', mimeType: 'image/png' }],
+      },
+    );
+
+    expect(result.content).toBe('ok');
+  });
+
   it('synthesizes openai-chat PiModel with reasoning=false for Qwen DashScope (#183)', async () => {
     getModelMock.mockReturnValue(undefined);
     completeSimpleMock.mockImplementationOnce(async (model) => {

packages/providers/src/index.ts

@@ -308,7 +308,11 @@ function synthesizeWireModel(
   wire: GenerateOptions['wire'],
   baseUrl: string | undefined,
 ): PiModel {
-  const supportsImageInput = wire === 'openai-codex-responses';
+  const supportsImageInput =
+    wire === 'anthropic' ||
+    wire === 'openai-chat' ||
+    wire === 'openai-responses' ||
+    wire === 'openai-codex-responses';
   const api =
     wire === 'anthropic'
       ? 'anthropic-messages'