知らない分野を、原文にたどりつくまで芋づる式に掘る Claude Code のスキル
Python
fix: validate source names and extract file boundaries
nsh1lf2d244efacfb
diff --git a/skills/imozuru/scripts/fetch.py b/skills/imozuru/scripts/fetch.pyindex 764852d..5327f54 100755--- a/skills/imozuru/scripts/fetch.py+++ b/skills/imozuru/scripts/fetch.py@@ -354,8 +354,21 @@ def save_verified(saved: Path, kept: Path, destination: Path) -> Path:return target+def valid_source_name(name: str) -> bool:+ """保存する原文の名前は、パスではなく単一のファイル名に限る。"""+ device = name.split(".", 1)[0].rstrip(" ").upper()+ return (bool(name.strip(" .")) and name == name.rstrip(" .")+ and not any(c in name for c in "/\\:")+ and not any(ord(c) < 32 or 127 <= ord(c) <= 159 for c in name)+ and device not in {"CON", "PRN", "AUX", "NUL", "CONIN$", "CONOUT$"}+ and not re.fullmatch(r"(?:COM|LPT)[1-9¹²³]", device))++def fetch_source(sources: Path, name: str, url: str, destination: Path) -> int:"""sources は今回の取得専用の一時置き場。取得・変換・検査中は保存済みの原文に触れない。"""+ if not valid_source_name(name):+ print(f"FAIL\t{name!r}\t原文の名前は単一のファイル名にする")+ return 1if X_POST.match(url):print(f"FAIL\t{name}\tX はログインの壁で本文が取れない(URL を map.md に残す)")@@ -455,6 +468,9 @@ def main() -> int:print(__doc__, file=sys.stderr)return 2sources, name, url = Path(sys.argv[1]), sys.argv[2], sys.argv[3]+ if not valid_source_name(name):+ print(f"FAIL\t{name!r}\t原文の名前は単一のファイル名にする")+ return 1sources.mkdir(parents=True, exist_ok=True)# 変換や FAIL の後始末は一時置き場だけで行う。同名の原文を検査前に上書きしない。# 保存先の中に作るので、成功時の replace はファイルシステムをまたがない。diff --git a/skills/imozuru/scripts/report.py b/skills/imozuru/scripts/report.pyindex 4eaffae..493b759 100644--- a/skills/imozuru/scripts/report.py+++ b/skills/imozuru/scripts/report.py@@ -25,7 +25,7 @@ import htmlimport jsonimport reimport sys-from pathlib import Path+from pathlib import Path, PureWindowsPath# 前後に見せる行数。抜き書きがどんな流れの中にあったかを、ページの中で開いて確かめられるように。CONTEXT_LINES = 6@@ -39,8 +39,18 @@ def norm(s: str) -> str:def verify(src_dir: Path, q: dict) -> tuple[bool, list[str], int, int]:"""抜き書きが原文の指定の行(前後2行の余裕つき)にそのまま在るか。前後の文脈も返す。"""- path = src_dir / q["file"]- if not path.exists():+ relative = Path(q["file"])+ if relative.is_absolute() or PureWindowsPath(q["file"]).is_absolute() or ".." in relative.parts:+ return False, [], 0, 0+ try:+ root = src_dir.resolve()+ topic = src_dir.parent.resolve()+ path = (src_dir / relative).resolve()+ except (OSError, RuntimeError, ValueError):+ return False, [], 0, 0+ if root == topic or not root.is_relative_to(topic):+ return False, [], 0, 0+ if not path.is_relative_to(root) or not path.is_file():return False, [], 0, 0# splitlines() はページ区切り(\f)も改行に数え、grep・sed と行番号がずれる。改行だけで数えるlines = path.read_text(encoding="utf-8", errors="replace").split("\n")@@ -50,8 +60,9 @@ def verify(src_dir: Path, q: dict) -> tuple[bool, list[str], int, int]:# 2段組の PDF は左右の段が1行に混ざり、何行かにまたがる文はそのままでは照合できない# (Yoshida 2008 で起きた)。抜き書きを「…」で区切れば、区切りごとに順に照らす。# 区切りはページにも「…」として残るので、省いたことが読む人に見える。- ok, pos = True, 0- for piece in [norm(p) for p in q["text"].split("…") if p.strip()]:+ pieces = [norm(p) for p in q["text"].split("…") if norm(p)]+ ok, pos = bool(pieces), 0+ for piece in pieces:at = region.find(piece, pos)if at < 0:ok = Falsediff --git a/tests/test_file_boundaries.py b/tests/test_file_boundaries.pynew file mode 100644index 0000000..85c849a--- /dev/null+++ b/tests/test_file_boundaries.py@@ -0,0 +1,109 @@+"""取得名と抜き書きの参照先が sources/ の境界を守ることを検査する。"""++import contextlib+import importlib.util+import io+import json+import sys+import tempfile+import unittest+from pathlib import Path+from unittest.mock import patch+++SCRIPTS = Path(__file__).resolve().parents[1] / "skills/imozuru/scripts"+++def load(name):+ spec = importlib.util.spec_from_file_location(f"boundary_{name}", SCRIPTS / f"{name}.py")+ module = importlib.util.module_from_spec(spec)+ spec.loader.exec_module(module)+ return module+++fetch = load("fetch")+report = load("report")+++class FileBoundaryTests(unittest.TestCase):+ def setUp(self):+ tmp = tempfile.TemporaryDirectory()+ self.addCleanup(tmp.cleanup)+ self.topic = Path(tmp.name) / "topic"+ self.sources = self.topic / "sources"+ self.sources.mkdir(parents=True)++ def quote(self, file, text="Original sentence"):+ return {"file": file, "line": 1, "text": text}++ def test_fetch_rejects_unsafe_names_before_side_effects(self):+ destination = self.topic / "saved"+ destination.mkdir()+ original = destination / "outside.md"+ original.write_text("Keep this source", encoding="utf-8")+ for name in ("", ".", "..", "...", " ", "../outside", "a/b", "a\\b",+ "/outside", "C:\\outside", "C:outside", "CON", "a\x00b", "a\nb"):+ with self.subTest(name=name), patch.object(fetch, "download") as download:+ with contextlib.redirect_stdout(io.StringIO()) as output:+ self.assertEqual(fetch.fetch_source(self.sources, name, "https://example.test/a", destination), 1)+ self.assertIn("FAIL\t", output.getvalue())+ download.assert_not_called()+ self.assertEqual(original.read_text(encoding="utf-8"), "Keep this source")+ self.assertEqual(list(self.sources.iterdir()), [])++ def test_fetch_accepts_author_year_unicode_name(self):+ name = "髙橋-2024-色彩の研究"+ body = b"Verified article text. " * 80+ with patch.object(fetch, "download", return_value=(body, "text/plain")):+ with contextlib.redirect_stdout(io.StringIO()):+ self.assertEqual(fetch.fetch_source(self.sources, name, "https://example.test/a", self.sources), 0)+ self.assertEqual((self.sources / f"{name}.md").read_bytes(), body)++ def test_report_rejects_outside_paths(self):+ outside = self.topic / "outside.md"+ outside.write_text("Original sentence", encoding="utf-8")+ for file in ("../outside.md", str(outside), "C:\\outside.md", "\x00"):+ with self.subTest(file=file):+ self.assertEqual(report.verify(self.sources, self.quote(file)), (False, [], 0, 0))++ def test_report_rejects_symlink_escapes(self):+ outside = self.topic / "outside.md"+ outside.write_text("Original sentence", encoding="utf-8")+ external = self.topic.parent / "external"+ external.mkdir()+ (external / "paper.md").write_text("Original sentence", encoding="utf-8")+ try:+ (self.sources / "link.md").symlink_to(outside)+ (self.topic / "linked-sources").symlink_to(external, target_is_directory=True)+ except (OSError, NotImplementedError) as exc:+ self.skipTest(f"Symlinks unavailable: {exc}")+ self.assertEqual(report.verify(self.sources, self.quote("link.md")), (False, [], 0, 0))+ self.assertEqual(report.verify(self.topic / "linked-sources", self.quote("paper.md")),+ (False, [], 0, 0))++ def test_report_rejects_empty_extracts_and_accepts_nested_source(self):+ nested = self.sources / "archive"+ nested.mkdir()+ (nested / "paper.md").write_text("Original sentence", encoding="utf-8")+ for text in ("", " \t\n ", "…", " … … "):+ with self.subTest(text=text):+ self.assertEqual(report.verify(self.sources, self.quote("archive/paper.md", text))[0], False)+ ok, context, _, _ = report.verify(self.sources, self.quote("archive/paper.md"))+ self.assertTrue(ok)+ self.assertIn("Original sentence", context)++ def test_failed_report_keeps_prior_content(self):+ (self.sources / "paper.md").write_text("Original sentence", encoding="utf-8")+ data = {"title": "Test", "question": "Q", "stopped": "Done", "claims": [+ {"claim": "Claim", "status": "付いた", "quotes": [self.quote("paper.md", "…")]}]}+ (self.topic / "extracts.json").write_text(json.dumps(data), encoding="utf-8")+ prior = self.topic / "report.html"+ prior.write_text("Previous report", encoding="utf-8")+ with patch.object(sys, "argv", ["report.py", str(self.topic)]), \+ contextlib.redirect_stdout(io.StringIO()):+ self.assertEqual(report.main(), 1)+ self.assertEqual(prior.read_text(encoding="utf-8"), "Previous report")+++if __name__ == "__main__":+ unittest.main()