fix(tools): normalize Unicode space family and minus sign in patch fuzzy matching

Port from anomalyco/opencode#38133/#38134 (patch Unicode matching corpus):
extend UNICODE_MAP with the Zs space-separator family (en/em quad, en/em/
three-per-em/four-per-em/six-per-em/figure/punctuation/thin/hair spaces,
narrow NBSP, medium mathematical space, CJK ideographic space) and the
Unicode minus sign U+2212.

Before: a file containing typographic spacing (French narrow NBSP, CJK
ideographic spaces, math minus) never matched a model's ASCII old_string
via the precise strategies — the edit only succeeded through the
similarity-based context_aware fallback, which (a) can pick the wrong
region (#54572 family) and (b) silently flattens the file's Unicode to
ASCII on replacement. After: these match at unicode_normalized (strategy
7), whose _preserve_unicode_in_replacement keeps the file's typographic
characters in unchanged spans.

All additions are 1:1 mappings, so the existing position-mapping and
preservation logic apply unchanged. Proven live before/after with a
multi-line probe; 59 fuzzy-match + 188 file-tools/patch/skill-manager
tests pass.

tests/tools/test_fuzzy_match.py

@@ -320,6 +320,58 @@ class TestUnicodeNormalized:
         assert new == "plain text there"
 
 
+class TestUnicodeSpaceAndMinusNormalized:
+    """Space-separator family + Unicode minus normalization.
+
+    Port of the anomalyco/opencode#38133 patch-matching corpus: files with
+    typographic spacing (en/em/thin spaces, narrow NBSP, CJK ideographic
+    space) or the Unicode minus sign must match a model's ASCII old_string
+    at the precise unicode_normalized strategy — not fall through to the
+    similarity-based context_aware fallback.
+    """
+
+    def test_unicode_minus_matched_and_preserved(self):
+        content = "offset = value \u2212 1\nprint(offset)\n"
+        new, count, strategy, err = fuzzy_find_and_replace(
+            content, "offset = value - 1", "offset = delta - 1"
+        )
+        assert count == 1, f"Expected match, got err={err}"
+        assert strategy == "unicode_normalized"
+        # The untouched minus keeps its Unicode form
+        assert "delta \u2212 1" in new, f"Got {new!r}"
+
+    def test_space_variants_match_at_unicode_strategy(self):
+        # en space, em space, thin space, narrow NBSP, medium math space,
+        # ideographic space, figure space, hair space
+        for space in ["\u2002", "\u2003", "\u2009", "\u202f",
+                      "\u205f", "\u3000", "\u2007", "\u200a"]:
+            content = f"# wait{space}30{space}seconds\nrun()\n"
+            new, count, strategy, err = fuzzy_find_and_replace(
+                content, "# wait 30 seconds", "# wait 60 seconds"
+            )
+            assert count == 1, (
+                f"space U+{ord(space):04X}: expected match, err={err}"
+            )
+            assert strategy == "unicode_normalized", (
+                f"space U+{ord(space):04X}: matched via {strategy}, "
+                "expected unicode_normalized"
+            )
+            # Unchanged spaces keep their typographic form; only the
+            # digits change.
+            assert f"60{space}seconds" in new, (
+                f"space U+{ord(space):04X}: got {new!r}"
+            )
+
+    def test_ideographic_space_cjk_line(self):
+        content = "標題\u3000第一章\nbody text\n"
+        new, count, strategy, err = fuzzy_find_and_replace(
+            content, "標題 第一章", "標題 第二章"
+        )
+        assert count == 1, f"Expected match, got err={err}"
+        assert strategy == "unicode_normalized"
+        assert "標題\u3000第二章" in new, f"Got {new!r}"
+
+
 class TestBlockAnchorThreshold:
     """Tests for the raised block_anchor threshold (Bug 4)."""
 

tools/fuzzy_match.py

@@ -38,6 +38,23 @@ UNICODE_MAP = {
     "\u2018": "'", "\u2019": "'",  # smart single quotes
     "\u2014": "--", "\u2013": "-", # em/en dashes
     "\u2026": "...", "\u00a0": " ", # ellipsis and non-breaking space
+    # Unicode minus sign — models type ASCII '-' for file content that uses
+    # the typographic minus (math/scientific docs).
+    "\u2212": "-",
+    # Space-separator family (Zs) beyond NBSP.  Files with typographic
+    # spacing (en/em/thin spaces, narrow NBSP in French text, ideographic
+    # space in CJK text) never match a model's ASCII-space old_string via
+    # the precise strategies, falling through to the similarity-based
+    # context_aware fallback — which can pick the wrong region and flattens
+    # the file's Unicode on replacement.  (anomalyco/opencode#38133 corpus)
+    "\u2000": " ", "\u2001": " ",  # en/em quad
+    "\u2002": " ", "\u2003": " ",  # en/em space
+    "\u2004": " ", "\u2005": " ", "\u2006": " ",  # three/four/six-per-em
+    "\u2007": " ", "\u2008": " ",  # figure/punctuation space
+    "\u2009": " ", "\u200a": " ",  # thin/hair space
+    "\u202f": " ",  # narrow no-break space
+    "\u205f": " ",  # medium mathematical space
+    "\u3000": " ",  # ideographic (CJK full-width) space
 }
 
 def _unicode_normalize(text: str) -> str: