xskill 0.3.0__py3-none-any.whl

xskill/__init__.py

@@ -0,0 +1,29 @@
+"""xskill — 从 AI Agent 执行轨迹自动蒸馏可复用 Skill。
+
+公开 SDK：
+    from xskill import XSkill, Skill, Trajectory, Evaluator
+    from xskill.types import (
+        WatchDir, SkillHit, TrajectoryHit,
+        EvalScore, Candidate, UxScoreResult,
+    )
+
+进阶（少数场景，例如单测直接拿子系统）：
+    from xskill import Registry, SkillRepo
+"""
+
+__version__ = "0.3.0"
+
+# 顶级公开面：4 个核心类
+from xskill.core import XSkill
+from xskill.entities.skill import Skill
+from xskill.entities.trajectory import Trajectory
+from xskill.entities.evaluator import Evaluator
+
+# 进阶：子系统类（不必常用）
+from xskill.entities.registry import Registry
+from xskill.entities.skill_repo import SkillRepo
+
+__all__ = [
+    "XSkill", "Skill", "Trajectory", "Evaluator",
+    "Registry", "SkillRepo",
+]

xskill/adapters.py

@@ -0,0 +1,183 @@
+"""
+adapters.py -- Trajectory submission and format adaptation layer
+=================================================================
+Convert various input formats to the standard xskill format
+(markdown + json metadata) and handle submission to the traj directory.
+"""
+
+import json
+import re
+from pathlib import Path
+from typing import Optional
+
+from xskill.config import get_traj_dir
+
+
+def generate_traj_id(traj_dir: Path = None) -> str:
+    """
+    Auto-generate a traj ID like ``traj_0301`` based on existing files
+    in *traj_dir*.  Scans for ``traj_*.md`` and picks max + 1.
+    """
+    traj_dir = traj_dir or get_traj_dir()
+    traj_dir.mkdir(parents=True, exist_ok=True)
+
+    existing_ids: list[int] = []
+    for f in traj_dir.glob("traj_*.md"):
+        m = re.match(r"traj_(\d+)", f.stem)
+        if m:
+            existing_ids.append(int(m.group(1)))
+
+    next_id = max(existing_ids) + 1 if existing_ids else 1
+    return f"traj_{next_id:04d}"
+
+
+def adapt_trajectory(
+    content: str,
+    format: str,
+    metadata: Optional[dict] = None,
+) -> tuple[str, dict]:
+    """
+    Convert various input formats to the standard xskill representation.
+
+    Supported *format* values:
+
+    - ``markdown`` -- passthrough; content is already ``traj_*.md`` format.
+    - ``json`` -- JSON object with fields like ``messages``, ``tool_calls``, etc.
+      Converted to a markdown trajectory.
+    - ``raw`` -- plain text; wrapped in a basic trajectory markdown template.
+
+    Returns ``(md_content, json_metadata)``.
+    """
+    metadata = metadata or {}
+
+    if format == "markdown":
+        return content, metadata
+
+    if format == "json":
+        return _adapt_json(content, metadata)
+
+    if format == "raw":
+        return _adapt_raw(content, metadata)
+
+    raise ValueError(f"unsupported trajectory format: {format!r}")
+
+
+# ------------------------------------------------------------------
+# Internal converters
+# ------------------------------------------------------------------
+
+def _adapt_json(content: str, metadata: dict) -> tuple[str, dict]:
+    """Convert a JSON trajectory to markdown + metadata."""
+    data = json.loads(content)
+
+    # Merge top-level keys (except messages/tool_calls) into metadata
+    meta = dict(metadata)
+    for key in ("model", "instance_id", "repo", "task", "result", "exit_status"):
+        if key in data and key not in meta:
+            meta[key] = data[key]
+
+    # Build markdown from messages / tool_calls
+    lines: list[str] = []
+    lines.append(f"# Trajectory")
+    if meta.get("instance_id"):
+        lines.append(f"\n**instance_id**: {meta['instance_id']}")
+    if meta.get("model"):
+        lines.append(f"**model**: {meta['model']}")
+    lines.append("")
+
+    messages = data.get("messages", [])
+    tool_calls = data.get("tool_calls", [])
+
+    for msg in messages:
+        role = msg.get("role", "unknown")
+        text = msg.get("content", "")
+        lines.append(f"## {role.capitalize()}")
+        lines.append("")
+        if isinstance(text, str):
+            lines.append(text)
+        elif isinstance(text, list):
+            # multi-part content
+            for part in text:
+                if isinstance(part, dict):
+                    lines.append(part.get("text", str(part)))
+                else:
+                    lines.append(str(part))
+        lines.append("")
+
+    if tool_calls:
+        lines.append("## Tool Calls")
+        lines.append("")
+        for tc in tool_calls:
+            name = tc.get("name", tc.get("function", {}).get("name", "unknown"))
+            args = tc.get("arguments", tc.get("function", {}).get("arguments", ""))
+            lines.append(f"### {name}")
+            lines.append("```")
+            lines.append(args if isinstance(args, str) else json.dumps(args, ensure_ascii=False))
+            lines.append("```")
+            if tc.get("output"):
+                lines.append(f"\n**output**:\n```\n{tc['output']}\n```")
+            lines.append("")
+
+    md_content = "\n".join(lines)
+    return md_content, meta
+
+
+def _adapt_raw(content: str, metadata: dict) -> tuple[str, dict]:
+    """Wrap plain text in a basic trajectory markdown template."""
+    lines = [
+        "# Trajectory",
+        "",
+        "## Raw Content",
+        "",
+        content,
+        "",
+    ]
+    md_content = "\n".join(lines)
+    return md_content, dict(metadata)
+
+
+# ------------------------------------------------------------------
+# Submission
+# ------------------------------------------------------------------
+
+def submit_trajectory(
+    content: str,
+    format: str = "markdown",
+    metadata: Optional[dict] = None,
+    traj_id: Optional[str] = None,
+    traj_dir: Optional[Path] = None,
+) -> dict:
+    """
+    Complete submission flow:
+
+    1. Resolve *traj_dir* (from param or ``get_traj_dir()``).
+    2. Generate *traj_id* if not provided.
+    3. Adapt the input format to standard markdown + JSON metadata.
+    4. Write ``traj_{id}.md`` and optionally ``traj_{id}.json``.
+    5. Return ``{"traj_id": ..., "path": ..., "status": "stored"}``.
+    """
+    traj_dir = Path(traj_dir) if traj_dir else get_traj_dir()
+    traj_dir.mkdir(parents=True, exist_ok=True)
+
+    if not traj_id:
+        traj_id = generate_traj_id(traj_dir)
+
+    md_content, json_metadata = adapt_trajectory(content, format, metadata)
+
+    # Write markdown
+    md_path = traj_dir / f"{traj_id}.md"
+    md_path.write_text(md_content, encoding="utf-8")
+
+    # Write JSON metadata if non-empty
+    if json_metadata:
+        json_path = traj_dir / f"{traj_id}.json"
+        json_path.write_text(
+            json.dumps(json_metadata, ensure_ascii=False, indent=2),
+            encoding="utf-8",
+        )
+
+    return {
+        "traj_id": traj_id,
+        "path": str(md_path),
+        "status": "stored",
+    }