feat: OpenMesh 基础平台与 MD/PDF 转换技能

- 后端: coworker 智能体框架, WS API, 文件上传, 附件处理
- 前端: Open WebUI, 文件全量走 upload API (含 MD/TXT/JSON 等文本类)
- 技能: md-to-office (pandoc + wkhtmltopdf)
- 修复: 上传文件路径丢失, Agent 搜索浪费, 输出文件跑到 uploads/
- 打包: PyInstaller one-dir, 预打包 pandoc/wkhtmltopdf/chromium

tests/test_attachments.py

@@ -0,0 +1,220 @@
+"""Tests for image/text attachments.
+
+Layered: (1) the pure content builder, (2) the pass-through assumption — an image attachment
+reaches the provider's `messages` byte-for-byte unmodified (a spy provider, no network),
+(3) persistence of list-content messages, and (4) an OPT-IN live vision call that proves the
+model actually reads the image (`COWORKER_LIVE_VISION=1`, key read from the SecretStore).
+"""
+
+from __future__ import annotations
+
+import base64
+import os
+import struct
+import zlib
+
+import pytest
+
+from coworker.attachments import build_user_content, content_to_text
+
+
+# -- a tiny solid-color PNG (stdlib) so tests need no fixtures -------------------
+def _solid_png(r: int, g: int, b: int, size: int = 32) -> bytes:
+    raw = bytearray()
+    row = bytes((r, g, b)) * size
+    for _ in range(size):
+        raw.append(0)
+        raw.extend(row)
+
+    def chunk(typ: bytes, data: bytes) -> bytes:
+        c = typ + data
+        return (
+            struct.pack(">I", len(data))
+            + c
+            + struct.pack(">I", zlib.crc32(c) & 0xFFFFFFFF)
+        )
+
+    return (
+        b"\x89PNG\r\n\x1a\n"
+        + chunk(
+            b"IHDR", struct.pack(">IIBBBBB", size, size, 8, 2, 0, 0, 0)
+        )  # 8-bit RGB
+        + chunk(b"IDAT", zlib.compress(bytes(raw), 9))
+        + chunk(b"IEND", b"")
+    )
+
+
+def _data_url(r: int, g: int, b: int) -> str:
+    return "data:image/png;base64," + base64.b64encode(_solid_png(r, g, b)).decode()
+
+
+# -- (1) builder ----------------------------------------------------------------
+def test_no_attachments_returns_plain_string():
+    assert build_user_content("hello", []) == "hello"
+    assert build_user_content("hello", None) == "hello"
+
+
+def test_image_attachment_becomes_image_url_part():
+    url = _data_url(220, 30, 30)
+    content = build_user_content("what is this?", [{"kind": "image", "data_url": url}])
+    assert isinstance(content, list)
+    assert content[0] == {"type": "text", "text": "what is this?"}
+    assert content[1] == {"type": "image_url", "image_url": {"url": url}}
+
+
+def test_text_attachment_is_inlined():
+    content = build_user_content(
+        "", [{"kind": "text", "name": "notes.md", "text": "# Title\nbody"}]
+    )
+    assert isinstance(content, list)
+    assert (
+        content[0]["type"] == "text"
+        and "notes.md" in content[0]["text"]
+        and "# Title" in content[0]["text"]
+    )
+
+
+def test_invalid_and_oversized_attachments_are_skipped():
+    bad = [
+        {"kind": "image", "data_url": "https://example.com/x.png"},  # not a data: URL
+        {
+            "kind": "image",
+            "data_url": "data:image/png;base64," + "A" * 12_000_001,
+        },  # too big
+        {"kind": "text", "text": ""},  # empty
+    ]
+    # only the leading text survives → falls back to the plain string
+    assert build_user_content("hi", bad) == "hi"
+
+
+def test_content_to_text_flattens_parts():
+    url = _data_url(0, 0, 0)
+    parts = build_user_content("look at this", [{"kind": "image", "data_url": url}])
+    assert content_to_text(parts) == "look at this [image]"
+    assert content_to_text("plain") == "plain"
+
+
+# -- (2) the assumption: image reaches the provider unmodified ------------------
+async def test_image_reaches_provider_unmodified():
+    from coworker.agents.chat import chat_agent
+    from coworker.agent import build_engine
+    from coworker.providers import AssistantTurn, ModelCapabilities, ProviderClient
+
+    class Spy(ProviderClient):
+        def __init__(self):
+            self.captured = None
+
+        def complete(self, *, model, messages, tools=None, **settings):
+            self.captured = [dict(m) for m in messages]
+            return AssistantTurn(text="ok", finish_reason="stop")
+
+        def capabilities(self, model):
+            return ModelCapabilities(vision=True)
+
+    spy = Spy()
+    engine = build_engine(agent=chat_agent(), model="gpt-4o", provider=spy)
+    url = _data_url(220, 30, 30)
+    content = build_user_content(
+        "describe the image", [{"kind": "image", "data_url": url}]
+    )
+
+    async for _ in engine.run(content):
+        pass
+
+    user_msgs = [m for m in (spy.captured or []) if m.get("role") == "user"]
+    assert user_msgs, "no user message reached the provider"
+    parts = user_msgs[-1]["content"]
+    assert isinstance(parts, list)
+    images = [p for p in parts if p.get("type") == "image_url"]
+    assert images and images[0]["image_url"]["url"] == url  # byte-for-byte intact
+
+
+# -- (3) persistence of list-content messages ----------------------------------
+def test_list_content_message_persists_and_titles(tmp_path):
+    from coworker.conversations import ConversationStore, title_from
+    from coworker.sessions import SessionRecord
+
+    url = _data_url(0, 128, 0)
+    msgs = [
+        {
+            "role": "user",
+            "content": build_user_content(
+                "review this diagram", [{"kind": "image", "data_url": url}]
+            ),
+        }
+    ]
+    store = ConversationStore(tmp_path)
+    rec = SessionRecord(
+        session_id="s1",
+        workspace=str(tmp_path),
+        model="gpt-4o",
+        mode="interactive",
+        messages=msgs,
+        agent="chat",
+    )
+    store.save(rec)
+    loaded = store.load("s1")
+    # the image survives the round-trip, and the title comes from the text part
+    assert loaded.messages[0]["content"][1]["image_url"]["url"] == url
+    assert title_from(msgs) == "review this diagram"
+
+
+# -- (4) live vision (opt-in) --------------------------------------------------
+@pytest.mark.skipif(
+    os.environ.get("COWORKER_LIVE_VISION") != "1",
+    reason="opt-in: real OpenAI vision call (set COWORKER_LIVE_VISION=1)",
+)
+def test_live_vision_model_reads_image():
+    from coworker.providers import OpenAIProvider
+    from coworker.secrets import SecretStore
+
+    provider = OpenAIProvider(default_model="gpt-4o", secrets=SecretStore())
+    content = build_user_content(
+        "What is the single dominant color of this image? Reply with one word.",
+        [{"kind": "image", "data_url": _data_url(220, 30, 30)}],
+    )
+    turn = provider.complete(
+        model="gpt-4o", messages=[{"role": "user", "content": content}]
+    )
+    assert "red" in (turn.text or "").lower(), f"model said: {turn.text!r}"
+
+
+def test_pdf_attachment_becomes_file_part():
+    url = "data:application/pdf;base64,JVBERi0xLjQ="
+    content = build_user_content(
+        "summarize this", [{"kind": "pdf", "name": "report.pdf", "data_url": url}]
+    )
+    assert content == [
+        {"type": "text", "text": "summarize this"},
+        {"type": "file", "file": {"filename": "report.pdf", "file_data": url}},
+    ]
+
+
+def test_pdf_attachment_invalid_or_oversized_skipped():
+    from coworker.attachments import MAX_PDF_CHARS
+
+    bad = [
+        {"kind": "pdf", "name": "x.pdf", "data_url": "data:image/png;base64,zz"},
+        {"kind": "pdf", "name": "x.pdf"},
+        {
+            "kind": "pdf",
+            "name": "big.pdf",
+            "data_url": "data:application/pdf;base64," + "A" * MAX_PDF_CHARS,
+        },
+    ]
+    assert build_user_content("hi", bad) == "hi"
+
+
+def test_content_to_text_renders_pdf_placeholder():
+    parts = [
+        {"type": "text", "text": "see attached"},
+        {
+            "type": "file",
+            "file": {
+                "filename": "report.pdf",
+                "file_data": "data:application/pdf;base64,x",
+            },
+        },
+    ]
+    assert content_to_text(parts) == "see attached [pdf]"
+    assert content_to_text(parts, image_placeholder="") == "see attached"