minimal-agent 0.9.0 → 0.10.0

package/skills/ui-design/scripts/resolve_source_input.py

@@ -0,0 +1,270 @@
+#!/usr/bin/env python3
+# -*- coding: utf-8 -*-
+"""
+resolve_source_input.py —— PPT 工作流素材入口解析。
+
+用途：
+  - inputs.source 明确给了路径：验证并返回该路径。
+  - inputs.source 为空，但用户自然语言里说了「下载目录里的某 Word 文件」：
+    在 Downloads / Desktop / cwd 内做受限搜索，找到唯一候选。
+  - 明确提到素材文件却找不到或多候选：写报告并退出非 0，避免下游伪装成纯 topic。
+
+只做受限目录搜索，不全盘扫描。
+"""
+
+import argparse
+import json
+import os
+import re
+import sys
+import unicodedata
+from dataclasses import dataclass
+
+
+SUPPORTED_EXTS = (
+    ".doc", ".docx", ".ppt", ".pptx", ".pdf", ".xlsx", ".xls", ".csv", ".tsv",
+    ".png", ".jpg", ".jpeg", ".gif", ".webp", ".bmp", ".tif", ".tiff",
+)
+WORD_EXTS = (".doc", ".docx")
+FILE_HINT_PATTERNS = (
+    r"[/\\]",
+    r"\.(doc|docx|ppt|pptx|pdf|xlsx|xls|csv|tsv|png|jpe?g|webp|gif)\b",
+    r"下载目录|Downloads|download",
+    r"桌面|Desktop|desktop",
+    r"Word\s*文件|word\s*文件|老\s*Word|老\s*word",
+    r"PPT\s*文件|ppt\s*文件|PowerPoint",
+    r"PDF\s*文件|pdf\s*文件",
+    r"源文件|素材文件|材料文件|文档文件",
+)
+STOP_WORDS = {
+    "用", "我", "电脑", "下载", "目录", "里面", "里的", "word", "Word", "WORD",
+    "文件", "作为", "基础", "生成", "一套", "专业", "中文", "PPT", "ppt",
+    "整体感", "不要", "卡片", "堆叠", "的", "和", "把",
+}
+
+
+@dataclass
+class Candidate:
+    path: str
+    score: int
+    reason: str
+
+
+def _norm(text: str) -> str:
+    text = unicodedata.normalize("NFKC", text or "")
+    return re.sub(r"\s+", "", text).lower()
+
+
+def _tokens(text: str) -> list[str]:
+    raw = unicodedata.normalize("NFKC", text or "")
+    parts = re.split(r"[\s/\\,，。；;:：\"'“”‘’()（）【】\[\]<>《》|]+", raw)
+    out: list[str] = []
+    for p in parts:
+        p = p.strip()
+        if not p or p in STOP_WORDS:
+            continue
+        if len(p) <= 1 and not re.search(r"[\u4e00-\u9fff]", p):
+            continue
+        out.append(p)
+    return out
+
+
+def mentions_source(topic: str, source: str) -> bool:
+    if source.strip():
+        return True
+    return any(re.search(pat, topic or "", re.I) for pat in FILE_HINT_PATTERNS)
+
+
+def search_roots(cwd: str, home: str, topic: str) -> list[str]:
+    roots = [cwd]
+    if "下载目录" in topic or "Downloads" in topic or "download" in topic.lower():
+        roots.insert(0, os.path.join(home, "Downloads"))
+    else:
+        roots.append(os.path.join(home, "Downloads"))
+    if "桌面" in topic or "Desktop" in topic or "desktop" in topic.lower():
+        roots.insert(0, os.path.join(home, "Desktop"))
+    else:
+        roots.append(os.path.join(home, "Desktop"))
+    seen = set()
+    out = []
+    for r in roots:
+        r = os.path.abspath(os.path.expanduser(r))
+        if r not in seen and os.path.isdir(r):
+            seen.add(r)
+            out.append(r)
+    return out
+
+
+def list_files(root: str, max_depth: int = 2) -> list[str]:
+    root = os.path.abspath(root)
+    files: list[str] = []
+    for cur, dirs, names in os.walk(root):
+        depth = os.path.relpath(cur, root).count(os.sep)
+        if depth >= max_depth:
+            dirs[:] = []
+        dirs[:] = [
+            d for d in dirs
+            if not d.startswith(".")
+            and d not in {
+                "node_modules", ".git", "vendor", "build", "out", "dist",
+                "coverage", ".cache", "tmp", "temp",
+            }
+        ]
+        for name in names:
+            if name.startswith("~$") or name.startswith("."):
+                continue
+            if name.lower().endswith(SUPPORTED_EXTS):
+                files.append(os.path.join(cur, name))
+    return sorted(files)
+
+
+def score_candidate(path: str, topic: str, source: str) -> Candidate | None:
+    base = os.path.basename(path)
+    ext = os.path.splitext(base)[1].lower()
+    norm_base = _norm(os.path.splitext(base)[0])
+    norm_topic = _norm(topic)
+    score = 0
+    reasons = []
+
+    if source.strip():
+        return Candidate(path, 10_000, "显式 source 路径")
+
+    toks = _tokens(topic)
+    for tok in toks:
+        nt = _norm(tok)
+        if not nt or nt in STOP_WORDS:
+            continue
+        if nt in norm_base:
+            add = max(2, min(20, len(nt)))
+            score += add
+            reasons.append(tok)
+
+    if ext in WORD_EXTS and re.search(r"word|doc|文档|文件", topic, re.I):
+        score += 20
+        reasons.append("Word 类型匹配")
+    elif ext in (".ppt", ".pptx") and re.search(r"ppt|幻灯|课件", topic, re.I):
+        score += 14
+        reasons.append("PPT 类型匹配")
+    elif ext in (".pdf",) and "pdf" in norm_topic:
+        score += 12
+        reasons.append("PDF 类型匹配")
+
+    # 长中文片段包含是强证据。
+    chinese_chunks = re.findall(r"[\u4e00-\u9fffA-Za-z0-9_]{4,}", topic)
+    for ch in chinese_chunks:
+        nch = _norm(ch)
+        if len(nch) >= 6 and nch in norm_base:
+            score += 30
+            reasons.append(ch)
+
+    if score <= 0:
+        return None
+    return Candidate(path, score, "、".join(reasons[:5]) or "文件名相似")
+
+
+def resolve_source(source: str, topic: str, cwd: str, home: str) -> dict:
+    source = (source or "").strip()
+    topic = topic or ""
+    explicit = bool(source)
+    wants_source = mentions_source(topic, source)
+
+    if explicit:
+        path = os.path.abspath(os.path.expanduser(source))
+        ok = os.path.exists(path)
+        return {
+            "ok": ok,
+            "explicit": True,
+            "requires_source": True,
+            "resolved_path": path if ok else None,
+            "candidates": [path] if ok else [],
+            "message": "显式 source 路径有效" if ok else f"显式 source 路径不存在：{path}",
+        }
+
+    candidates: list[Candidate] = []
+    seen_paths = set()
+    explicit_dir_hint = any(x in topic for x in ("下载目录", "Downloads", "桌面", "Desktop"))
+    if wants_source:
+        for root in search_roots(cwd, home, topic):
+            before = len(candidates)
+            for f in list_files(root):
+                real = os.path.realpath(f)
+                if real in seen_paths:
+                    continue
+                seen_paths.add(real)
+                cand = score_candidate(f, topic, source)
+                if cand:
+                    candidates.append(cand)
+            # 用户明确说了下载目录/桌面时，命中该优先目录即可停止，
+            # 避免继续扫大型仓库 cwd 导致入口阶段卡顿。
+            if explicit_dir_hint and len(candidates) > before:
+                break
+    candidates.sort(key=lambda c: (-c.score, len(c.path), c.path))
+
+    if not wants_source:
+        return {
+            "ok": True,
+            "explicit": False,
+            "requires_source": False,
+            "resolved_path": None,
+            "candidates": [],
+            "message": "未检测到明确素材文件指令，允许扫描 cwd 或纯 topic 路径",
+        }
+
+    if not candidates:
+        return {
+            "ok": False,
+            "explicit": False,
+            "requires_source": True,
+            "resolved_path": None,
+            "candidates": [],
+            "message": "用户提到素材文件，但在 Downloads/Desktop/cwd 未找到匹配候选",
+        }
+
+    top = candidates[0]
+    tied = [c for c in candidates if c.score == top.score]
+    if len(tied) > 1:
+        return {
+            "ok": False,
+            "explicit": False,
+            "requires_source": True,
+            "resolved_path": None,
+            "candidates": [{"path": c.path, "score": c.score, "reason": c.reason} for c in tied[:10]],
+            "message": "用户提到素材文件，但存在多个同分候选，请显式传 --input source",
+        }
+
+    return {
+        "ok": True,
+        "explicit": False,
+        "requires_source": True,
+        "resolved_path": top.path,
+        "candidates": [{"path": c.path, "score": c.score, "reason": c.reason} for c in candidates[:10]],
+        "message": f"按自然语言解析到素材：{top.path}",
+    }
+
+
+def main() -> int:
+    parser = argparse.ArgumentParser(description="解析 PPT 工作流输入里的素材路径")
+    parser.add_argument("--source", default="", help="用户显式传入的素材路径")
+    parser.add_argument("--topic", default="", help="用户原始自然语言指令")
+    parser.add_argument("--cwd", default=".", help="工作流 cwd")
+    parser.add_argument("--out", default="./build/source_resolution.json", help="解析报告输出路径")
+    parser.add_argument("--home", default=os.path.expanduser("~"), help="测试用 home 覆盖")
+    args = parser.parse_args()
+
+    result = resolve_source(args.source, args.topic, args.cwd, args.home)
+    os.makedirs(os.path.dirname(os.path.abspath(args.out)), exist_ok=True)
+    with open(args.out, "w", encoding="utf-8") as f:
+        json.dump(result, f, ensure_ascii=False, indent=2)
+    print(result["message"])
+    if result.get("ok"):
+        if result.get("resolved_path"):
+            print(f"RESOLVED_SOURCE={result['resolved_path']}")
+        return 0
+    print("SOURCE_RESOLUTION_FAILED", file=sys.stderr)
+    for c in result.get("candidates", []):
+        print(f"- {c}", file=sys.stderr)
+    return 2
+
+
+if __name__ == "__main__":
+    sys.exit(main())

package/skills/ui-design/scripts/resolve_source_input_test.py

@@ -0,0 +1,98 @@
+#!/usr/bin/env python3
+# -*- coding: utf-8 -*-
+
+import json
+import os
+import subprocess
+import sys
+import tempfile
+import unittest
+
+sys.path.insert(0, os.path.dirname(os.path.abspath(__file__)))
+import resolve_source_input as rs  # noqa: E402
+
+
+class TestResolveSourceInput(unittest.TestCase):
+    def test_explicit_absolute_path_wins(self):
+        tmp = tempfile.mkdtemp(prefix="resolve-source-")
+        doc = os.path.join(tmp, "材料.docx")
+        open(doc, "wb").close()
+        out = rs.resolve_source(doc, "随便做个 PPT", tmp, tmp)
+        self.assertTrue(out["ok"])
+        self.assertTrue(out["explicit"])
+        self.assertEqual(out["resolved_path"], doc)
+
+    def test_downloads_chinese_word_hint_finds_doc(self):
+        home = tempfile.mkdtemp(prefix="resolve-home-")
+        downloads = os.path.join(home, "Downloads")
+        os.makedirs(downloads)
+        doc = os.path.join(downloads, "航空器制造组装360度全景影像档案系统方案  20250328_副本.doc")
+        open(doc, "wb").close()
+        topic = "用我电脑下载目录里面的航空器制造组装360度全景影像档案系统方案 20250328_副本 Word 文件作为基础生成 PPT"
+        out = rs.resolve_source("", topic, tempfile.mkdtemp(prefix="resolve-cwd-"), home)
+        self.assertTrue(out["ok"], out)
+        self.assertTrue(out["requires_source"])
+        self.assertEqual(out["resolved_path"], doc)
+
+    def test_multiple_candidates_do_not_guess(self):
+        home = tempfile.mkdtemp(prefix="resolve-home-")
+        downloads = os.path.join(home, "Downloads")
+        os.makedirs(downloads)
+        open(os.path.join(downloads, "方案.doc"), "wb").close()
+        open(os.path.join(downloads, "方案.docx"), "wb").close()
+        out = rs.resolve_source("", "用下载目录里的方案 Word 文件做 PPT", tempfile.mkdtemp(), home)
+        self.assertFalse(out["ok"])
+        self.assertIn("多个", out["message"])
+
+    def test_missing_mentioned_source_fails(self):
+        home = tempfile.mkdtemp(prefix="resolve-home-")
+        os.makedirs(os.path.join(home, "Downloads"))
+        out = rs.resolve_source("", "用下载目录里的不存在 Word 文件做 PPT", tempfile.mkdtemp(), home)
+        self.assertFalse(out["ok"])
+        self.assertTrue(out["requires_source"])
+
+    def test_generic_material_topic_does_not_force_file_lookup(self):
+        out = rs.resolve_source(
+            "",
+            "帮我做一套关于素材管理平台的 PPT，不要联网",
+            tempfile.mkdtemp(prefix="resolve-cwd-"),
+            tempfile.mkdtemp(prefix="resolve-home-"),
+        )
+        self.assertTrue(out["ok"], out)
+        self.assertFalse(out["requires_source"])
+        self.assertIsNone(out["resolved_path"])
+
+    def test_cli_writes_report(self):
+        tmp = tempfile.mkdtemp(prefix="resolve-cli-")
+        home = os.path.join(tmp, "home")
+        downloads = os.path.join(home, "Downloads")
+        os.makedirs(downloads)
+        doc = os.path.join(downloads, "项目方案.doc")
+        open(doc, "wb").close()
+        report = os.path.join(tmp, "source_resolution.json")
+        script = os.path.join(os.path.dirname(os.path.abspath(__file__)), "resolve_source_input.py")
+        res = subprocess.run(
+            [
+                sys.executable,
+                script,
+                "--topic",
+                "用下载目录里面的项目方案 Word 文件做 PPT",
+                "--cwd",
+                tmp,
+                "--home",
+                home,
+                "--out",
+                report,
+            ],
+            text=True,
+            stdout=subprocess.PIPE,
+            stderr=subprocess.PIPE,
+        )
+        self.assertEqual(res.returncode, 0, res.stderr)
+        with open(report, encoding="utf-8") as f:
+            data = json.load(f)
+        self.assertEqual(data["resolved_path"], doc)
+
+
+if __name__ == "__main__":
+    unittest.main(verbosity=2)