fix: Strip ANSI escapes before parsing token counts and reported models

ringer.py

@@ -9430,6 +9430,10 @@ def parse_env_file(path: Path) -> dict[str, str]:
 
 
 def parse_token_count(text: str, token_regex: str | None = DEFAULT_TOKEN_REGEX) -> int | None:
+    # Engines colorize stdout when FORCE_COLOR is set (Claude Code, most CI),
+    # leaving escape bytes between the label and the number that no practical
+    # regex bridges — strip them once here so every regex sees plain text.
+    text = ANSI_RE.sub("", text)
     if token_regex:
         matches = list(re.finditer(token_regex, text, flags=re.IGNORECASE))
         for match in reversed(matches):
@@ -9454,6 +9458,9 @@ def parse_token_count(text: str, token_regex: str | None = DEFAULT_TOKEN_REGEX)
 def parse_reported_model(text: str, model_report_regex: str | None) -> str | None:
     if not model_report_regex:
         return None
+    # Same reason as parse_token_count: a colorized "model:" header breaks
+    # the ^model: anchor unless the ANSI escapes are stripped first.
+    text = ANSI_RE.sub("", text)
     match = re.search(model_report_regex, text, flags=re.IGNORECASE)
     if match is None or match.lastindex is None:
         return None

tests/test_identity_evidence.py

@@ -105,6 +105,14 @@ def test_codex_report_regex_captures_synthetic_header(self) -> None:
         output = "OpenAI Codex v0.144.0\n--------\nmodel: gpt-5.6-sol\nprovider: openai\n"
         self.assertEqual("gpt-5.6-sol", parse_reported_model(output, engine.model_report_regex))
 
+    def test_codex_report_regex_captures_ansi_colorized_header(self) -> None:
+        # Real bytes from codex exec run with FORCE_COLOR set (as Claude Code
+        # and most CI systems do): the header is bold-wrapped, so the ^model:
+        # anchor never matches without stripping ANSI first.
+        engine = load_engines(None)["codex"]
+        output = "OpenAI Codex v0.146.0\n--------\n\x1b[1mmodel:\x1b[0m gpt-5.6-sol\nprovider: openai\n"
+        self.assertEqual("gpt-5.6-sol", parse_reported_model(output, engine.model_report_regex))
+
     def test_reported_model_wins_and_resolved_model_is_fallback(self) -> None:
         rows = self.log_attempts(
             WorkerResult(0, False, 12, reported_model="gpt-5.7"),

tests/test_ringer.py

@@ -595,6 +595,19 @@ def test_token_count_parser_accepts_colon_and_newline_formats(self) -> None:
         self.assertEqual(ringer.parse_token_count("tokens used: 1,234", r"tokens\s+used\s*:?\s*([0-9][0-9,]*)"), 1234)
         self.assertEqual(ringer.parse_token_count("tokens used\n5,678", r"tokens\s+used\s*:?\s*([0-9][0-9,]*)"), 5678)
 
+    def test_token_count_parser_sees_through_ansi_colorized_output(self) -> None:
+        # Real bytes from codex exec run with FORCE_COLOR set (as Claude Code
+        # and most CI systems do): the label is dim-wrapped and the number is
+        # on the following line, so the escape sequences sit between "used"
+        # and the digits and an ANSI-blind \s* cannot bridge them.
+        colorized = "\x1b[2mtokens used\x1b[0m\n17,270\n"
+        self.assertEqual(ringer.parse_token_count(colorized, ringer.DEFAULT_TOKEN_REGEX), 17270)
+        # User-supplied custom regexes get the same stripping — the fix is in
+        # the parser, not the shipped pattern.
+        self.assertEqual(ringer.parse_token_count(colorized, r"tokens\s+used\s*:?\s*([0-9][0-9,]*)"), 17270)
+        # Uncolorized output keeps parsing exactly as before.
+        self.assertEqual(ringer.parse_token_count("tokens used: 1,234", ringer.DEFAULT_TOKEN_REGEX), 1234)
+
 
 if __name__ == "__main__":
     unittest.main()