feat(cursor): Windows Cursor 适配与导入脚本

examples/config.yaml.example

@@ -23,9 +23,10 @@ llm:
 # ===== Embedding（向量检索用）=====
 embedding:
   base_url: https://ark.cn-beijing.volces.com/api/v3
-  model:    doubao-embedding-vision-251215
+  model:    doubao-embedding-vision-251215   # 或 doubao-embedding-large-text-250515
   api_key:  PUT_YOUR_EMBEDDING_API_KEY_HERE
-  dim:      0                      # 0 = 自动探测
+  dim:      0                      # 0 = 自动探测；large-text 可写 2048
+  # api: openai | multimodal      # 可选；默认 vision 模型→multimodal，text 模型→/embeddings
 
 # ===== L2 沙箱评估（SWE-bench docker 跑 A/B/C；单测 / 评估精度脚本用）=====
 sandbox:

scripts/cursor_import.py

@@ -0,0 +1,93 @@
+#!/usr/bin/env python3
+"""Import Cursor agent-transcripts (*.jsonl) into xskill watch dir as traj_*.md."""
+
+from __future__ import annotations
+
+import argparse
+import json
+import sys
+from pathlib import Path
+
+from xskill.adapters import submit_trajectory
+
+
+def _jsonl_to_markdown(jsonl_path: Path) -> str:
+    lines: list[str] = [
+        "# Cursor Agent Trajectory",
+        "",
+        f"**source_file**: {jsonl_path}",
+        "",
+    ]
+    for raw in jsonl_path.read_text(encoding="utf-8", errors="ignore").splitlines():
+        raw = raw.strip()
+        if not raw:
+            continue
+        ev = json.loads(raw)
+        role = ev.get("role", "unknown")
+        msg = ev.get("message") or {}
+        parts = msg.get("content") or []
+        chunks: list[str] = []
+        for p in parts:
+            if not isinstance(p, dict):
+                continue
+            if p.get("type") == "text" and p.get("text"):
+                chunks.append(str(p["text"]))
+            elif p.get("type") == "tool_use":
+                name = p.get("name", "tool")
+                chunks.append(f"[tool_use: {name}]")
+        body = "\n".join(chunks).strip()
+        if not body:
+            continue
+        lines.append(f"## {str(role).capitalize()}")
+        lines.append("")
+        lines.append(body)
+        lines.append("")
+    return "\n".join(lines)
+
+
+def main() -> int:
+    p = argparse.ArgumentParser(description=__doc__)
+    p.add_argument(
+        "--src",
+        type=Path,
+        default=Path.home()
+        / ".cursor/projects/c-yzj-entrepreneurship-XSKILL-xskill/agent-transcripts",
+        help="Cursor agent-transcripts root (searched recursively for *.jsonl)",
+    )
+    p.add_argument(
+        "--out",
+        type=Path,
+        default=Path.home() / ".xskill/cursor_import",
+        help="xskill watch directory (traj_*.md output)",
+    )
+    args = p.parse_args()
+    src = args.src.expanduser().resolve()
+    out = args.out.expanduser().resolve()
+    out.mkdir(parents=True, exist_ok=True)
+
+    jsonls = sorted(src.rglob("*.jsonl"))
+    if not jsonls:
+        print(f"no *.jsonl under {src}", file=sys.stderr)
+        return 1
+
+    for jsonl in jsonls:
+        md = _jsonl_to_markdown(jsonl)
+        sid = jsonl.stem
+        result = submit_trajectory(
+            content=md,
+            format="markdown",
+            metadata={
+                "source": "cursor",
+                "ecosystem": "cursor",
+                "session_id": sid,
+                "source_jsonl": str(jsonl),
+            },
+            traj_id=f"traj_cursor_{sid[:8]}",
+            traj_dir=out,
+        )
+        print(f"imported {jsonl.name} -> {result['path']}")
+    return 0
+
+
+if __name__ == "__main__":
+    raise SystemExit(main())

scripts/cursor_setup.ps1

@@ -0,0 +1,52 @@
+# xskill + Cursor one-shot setup (dirs, junction, import, registry)
+# Run from repo root: powershell -ExecutionPolicy Bypass -File scripts\cursor_setup.ps1
+
+$ErrorActionPreference = "Stop"
+$RepoRoot = Split-Path $PSScriptRoot -Parent
+
+$XskillHome = Join-Path $env:USERPROFILE ".xskill"
+$SkillStore = Join-Path $XskillHome "skill"
+$CursorImport = Join-Path $XskillHome "cursor_import"
+$CursorSkills = Join-Path $env:USERPROFILE ".cursor\skills"
+$ConfigPath = Join-Path $XskillHome "config.yaml"
+$ExampleConfig = Join-Path $RepoRoot "examples\config.yaml.example"
+$VenvPython = Join-Path $RepoRoot ".venv\Scripts\python.exe"
+$XskillExe = Join-Path $RepoRoot ".venv\Scripts\xskill.exe"
+
+Write-Host "`n[Step 1] Create directories"
+New-Item -ItemType Directory -Force -Path $XskillHome, $SkillStore, $CursorImport | Out-Null
+New-Item -ItemType Directory -Force -Path (Join-Path $env:USERPROFILE ".cursor") | Out-Null
+
+Write-Host "`n[Step 2] Copy config.yaml if missing"
+if (-not (Test-Path $ConfigPath)) {
+    Copy-Item $ExampleConfig $ConfigPath
+    Write-Host "Edit $ConfigPath and set llm.api_key / embedding.api_key"
+}
+
+Write-Host "`n[Step 3] Junction: .cursor\skills -> .xskill\skill"
+if (Test-Path $CursorSkills) {
+    $item = Get-Item $CursorSkills -Force
+    $reparse = ($item.Attributes -band [IO.FileAttributes]::ReparsePoint) -ne 0
+    if ($reparse) {
+        Write-Host "Already linked: $CursorSkills"
+    } else {
+        Write-Warning "$CursorSkills exists and is not a junction; remove it manually first."
+    }
+} else {
+    cmd /c "mklink /J `"$CursorSkills`" `"$SkillStore`""
+}
+
+Write-Host "`n[Step 4] pip install -e .[dev]"
+Set-Location $RepoRoot
+if (-not (Test-Path $VenvPython)) { python -m venv .venv }
+& $VenvPython -m pip install -q -e ".[dev]"
+
+Write-Host "`n[Step 5] Import Cursor agent-transcripts"
+& $VenvPython (Join-Path $RepoRoot "scripts\cursor_import.py")
+
+Write-Host "`n[Step 6] registry add cursor_import"
+& $XskillExe registry add $CursorImport --label cursor_import
+& $XskillExe registry list
+
+Write-Host "`nDone. Next: edit config.yaml keys, then:"
+Write-Host "  .\.venv\Scripts\xskill.exe serve --host 127.0.0.1 --port 8000"

src/xskill/config.py

@@ -42,7 +42,7 @@ def load_config(path: Optional[Path] = None) -> dict:
             f"xskill config not found: {cfg_path}\n"
             f"Create it manually (see docs)."
         )
-    with open(cfg_path) as f:
+    with open(cfg_path, encoding="utf-8") as f:
         _config = yaml.safe_load(f) or {}
     if not _config.get("llm", {}).get("api_key"):
         raise KeyError(f"llm.api_key missing in {cfg_path}")

src/xskill/git_lock.py

@@ -44,8 +44,19 @@
 
 
 def run_git(args: list[str], cwd: str) -> tuple[int, str, str]:
-    r = subprocess.run(["git"] + args, cwd=cwd, capture_output=True, text=True)
-    return r.returncode, r.stdout.strip(), r.stderr.strip()
+    """Run git in *cwd*; always decode UTF-8 (Windows 默认 GBK 会在 git 输出含非 ASCII 时炸).
+
+    subprocess 在解码失败时可能把 stdout/stderr 置为 None；调用方统一当空串处理。
+    """
+    r = subprocess.run(
+        ["git"] + args,
+        cwd=cwd,
+        capture_output=True,
+        text=True,
+        encoding="utf-8",
+        errors="replace",
+    )
+    return r.returncode, (r.stdout or "").strip(), (r.stderr or "").strip()
 
 
 def init_skill_repo_on_baby(skill_dir: str, name: str, description: str) -> None:

src/xskill/llm_client.py

@@ -18,9 +18,12 @@
 
 import os, json, logging, time
 from dataclasses import dataclass, field
+from typing import Literal
 
 import numpy as np
 
+EmbedApiStyle = Literal["multimodal", "openai"]
+
 logger = logging.getLogger(__name__)
 
 
@@ -143,12 +146,29 @@ def __repr__(self):
 # Embedding Client
 # ═══════════════════════════════════════════════════════════════════
 
+def _resolve_embed_api_style(cfg: dict, model: str) -> EmbedApiStyle:
+    """ARK 有两套 embedding 路径：
+
+    - ``/embeddings`` — OpenAI 兼容，用于 ``doubao-embedding-large-text-*`` 等纯文本模型
+    - ``/embeddings/multimodal`` — 用于 ``doubao-embedding-vision-*`` 等多模态模型
+
+    可在 config 里显式写 ``embedding.api: openai | multimodal``；否则按模型名推断。
+    """
+    explicit = (cfg.get("api") or cfg.get("api_style") or "").strip().lower()
+    if explicit in ("multimodal", "openai"):
+        return explicit  # type: ignore[return-value]
+    if "vision" in model.lower():
+        return "multimodal"
+    return "openai"
+
+
 @dataclass
 class EmbedClient:
     base_url: str
     model: str
     api_key: str
     dim: int = 0  # 0 = 未探测
+    api_style: EmbedApiStyle = "openai"
     _client: object = field(default=None, repr=False)
 
     @classmethod
@@ -163,7 +183,10 @@ def from_config(cls, cfg: dict) -> "EmbedClient":
         dim = cfg.get("dim", 0)
         if not base_url or not model:
             raise ValueError("embedding.base_url 和 embedding.model 必须配置")
-        inst = cls(base_url=base_url, model=model, api_key=api_key, dim=dim)
+        api_style = _resolve_embed_api_style(cfg, model)
+        inst = cls(
+            base_url=base_url, model=model, api_key=api_key, dim=dim, api_style=api_style,
+        )
         return inst
 
     def _get_session(self):
@@ -175,26 +198,50 @@ def _get_session(self):
                 logger.warning("T2S_SSL_VERIFY=false → Embedding HTTPS 证书验证已关闭")
         return self._client
 
-    def _call_api_single(self, text: str) -> list[float]:
-        """调用 ARK multimodal embedding 接口（单条）"""
+    def _post_json(self, path: str, body: dict) -> dict:
         session = self._get_session()
-        url = f"{self.base_url}/embeddings/multimodal"
+        url = f"{self.base_url}{path}"
         headers = {
             "Content-Type": "application/json",
             "Authorization": f"Bearer {self.api_key}",
         }
-        body = {"model": self.model, "input": [{"type": "text", "text": text}]}
         resp = session.post(url, json=body, headers=headers)
         resp.raise_for_status()
-        data = resp.json()
+        return resp.json()
+
+    def _call_api_multimodal(self, text: str) -> list[float]:
+        """ARK multimodal：``doubao-embedding-vision-*`` 等"""
+        data = self._post_json(
+            "/embeddings/multimodal",
+            {"model": self.model, "input": [{"type": "text", "text": text}]},
+        )
         return data["data"]["embedding"]
 
+    def _call_api_openai(self, text: str) -> list[float]:
+        """ARK / OpenAI 兼容：``POST /embeddings``，``doubao-embedding-large-text-*`` 等"""
+        data = self._post_json(
+            "/embeddings",
+            {"model": self.model, "input": text},
+        )
+        items = data.get("data") or []
+        if not items:
+            raise ValueError(f"embedding response missing data: {data!r}")
+        return items[0]["embedding"]
+
+    def _call_api_single(self, text: str) -> list[float]:
+        if self.api_style == "multimodal":
+            return self._call_api_multimodal(text)
+        return self._call_api_openai(text)
+
     def probe_dim(self) -> int:
         """发送测试文本，探测 embedding 维度"""
         if self.dim > 0:
             return self.dim
 
-        logger.info(f"探测 embedding 维度: {self.model} @ {self.base_url}")
+        logger.info(
+            "探测 embedding 维度: %s @ %s (api=%s)",
+            self.model, self.base_url, self.api_style,
+        )
         vec = self._call_api_single("hello")
         self.dim = len(vec)
         logger.info(f"探测完成: dim={self.dim}")
@@ -208,7 +255,7 @@ def encode(self, text: str) -> np.ndarray:
         return vec
 
     def encode_batch(self, texts: list[str]) -> np.ndarray:
-        """批量文本 → (n, dim) 矩阵，逐条调用 multimodal 端点"""
+        """批量文本 → (n, dim) 矩阵，逐条调用 embedding 端点"""
         from tqdm import tqdm
         all_vecs = []
 
@@ -229,7 +276,10 @@ def encode_batch(self, texts: list[str]) -> np.ndarray:
         return result
 
     def __repr__(self):
-        return f"EmbedClient(base_url={self.base_url}, model={self.model}, dim={self.dim})"
+        return (
+            f"EmbedClient(base_url={self.base_url}, model={self.model}, "
+            f"dim={self.dim}, api_style={self.api_style})"
+        )
 
 
 # ═══════════════════════════════════════════════════════════════════

tests/test_llm_client.py

@@ -15,7 +15,7 @@
 
 import pytest
 
-from xskill.llm_client import LLMClient
+from xskill.llm_client import LLMClient, EmbedClient, _resolve_embed_api_style
 
 
 class TestLLMClientDefaults:
@@ -65,3 +65,21 @@ def test_missing_base_url_raises(self):
     def test_missing_model_raises(self):
         with pytest.raises(ValueError):
             LLMClient.from_config({"base_url": "http://x", "api_key": "k"})
+
+
+class TestEmbedApiStyle:
+    def test_text_model_defaults_openai(self):
+        assert _resolve_embed_api_style({}, "doubao-embedding-large-text-250515") == "openai"
+
+    def test_vision_model_defaults_multimodal(self):
+        assert _resolve_embed_api_style({}, "doubao-embedding-vision-251215") == "multimodal"
+
+    def test_explicit_api_override(self):
+        cfg = {"api": "multimodal"}
+        assert _resolve_embed_api_style(cfg, "doubao-embedding-large-text-250515") == "multimodal"
+
+    def test_from_config_sets_api_style(self):
+        c = EmbedClient.from_config({
+            "base_url": "http://x", "model": "doubao-embedding-large-text-250515", "api_key": "k",
+        })
+        assert c.api_style == "openai"