fix: close reviewed tool identity and credential redaction edge cases

agent/cli/ui/rail.py

@@ -473,9 +473,10 @@ class RailRunDashboard:
             for step in self.steps:
                 if step.call_id == raw_call_id:
                     return step
-        for step in reversed(self.steps):
-            if step.status == "active" and (not raw_tool or step.tool == raw_tool):
-                return step
+        else:
+            for step in reversed(self.steps):
+                if step.status == "active" and (not raw_tool or step.tool == raw_tool):
+                    return step
         step = RailStep(title=_tool_title(raw_tool or "tool", None), tool=raw_tool, call_id=raw_call_id)
         self.steps.append(step)
         self.steps = self.steps[-10:]

agent/src/tools/redaction.py

@@ -510,11 +510,16 @@ _TEXT_PRIVATE_KEY_PATTERN = re.compile(
 #: both leaked it. The value runs to the end of the line or the closing quote of
 #: a quoted shell argument. An already-redacted value is skipped, which keeps the
 #: substitution idempotent.
+_TEXT_QUOTED_COOKIE_HEADER_PATTERN = re.compile(
+    r"(?P<quote>[\"'])(?P<head>(?:set-)?cookie[ \t]*:)(?P<sp>[ \t]*)"
+    r"(?:(?:\\.)|(?!(?P=quote))[^\r\n\\])*(?P=quote)",
+    re.IGNORECASE,
+)
 _TEXT_COOKIE_HEADER_PATTERN = re.compile(
     r"(?P<head>(?<![\w-])(?:set-)?cookie\s*:)"
     r"(?P<sp>[ \t]*)(?![ \t])"
     r"(?!\[redacted\])"
-    r"[^\r\n\"']+",
+    r"[^\r\n]+",
     re.IGNORECASE,
 )
 
@@ -589,6 +594,9 @@ def redact_text(text: object) -> str:
     if not s:
         return s
     s = _TEXT_PRIVATE_KEY_PATTERN.sub(_REDACTED, s)
+    s = _TEXT_QUOTED_COOKIE_HEADER_PATTERN.sub(
+        r"\g<quote>\g<head>\g<sp>" + _REDACTED + r"\g<quote>", s,
+    )
     s = _TEXT_COOKIE_HEADER_PATTERN.sub(r"\g<head>\g<sp>" + _REDACTED, s)
     s = _TEXT_BEARER_PATTERN.sub(r"\g<scheme> " + _REDACTED, s)
     s = _TEXT_AUTH_SCHEME_PATTERN.sub(r"\g<head>\g<scheme> " + _REDACTED, s)

agent/src/tools/skill_writer_tool.py

@@ -69,7 +69,7 @@ def _with_forced_frontmatter_name(content: str, slug: str) -> str:
     new_lines = []
     replaced = False
     for line in header_lines:
-        if line.strip().startswith("name:"):
+        if line.partition(":")[0].strip() == "name":
             new_lines.append(f"name: {slug}")
             replaced = True
         else:

agent/src/tools/trade_journal_parsers.py

@@ -26,11 +26,7 @@ _CURRENCY_TOKEN_RE = re.compile(
 
 # ISO codes of the fiat currencies a broker export can quote as a pair
 # (EUR/USD); a pair with two fiat legs is forex, never crypto.
-_FIAT_CODES = frozenset({
-    "USD", "EUR", "GBP", "JPY", "CHF", "AUD", "CAD", "NZD", "CNY", "CNH", "HKD",
-    "SGD", "SEK", "NOK", "DKK", "MXN", "ZAR", "INR", "KRW", "TRY", "PLN", "CZK",
-    "HUF", "ILS", "THB",
-})
+from src.market_data import FIAT_CODES as _FIAT_CODES
 
 FormatName = str  # "tonghuashun" | "eastmoney" | "futu" | "generic" | "unknown"
 

agent/tests/test_cli_rail_dashboard.py

@@ -28,6 +28,12 @@ class TestConcurrentToolCallPairing:
     it actually belongs to.
     """
 
+    def test_missing_call_does_not_complete_another_active_call(self, dashboard):
+        dashboard.handle_event("tool_call", {"tool": "backtest", "call_id": "active"})
+        dashboard.handle_event("tool_result", {"tool": "backtest", "call_id": "evicted", "status": "error"})
+        assert next(step for step in dashboard.steps if step.call_id == "active").status == "active"
+        assert next(step for step in dashboard.steps if step.call_id == "evicted").status == "error"
+
     def test_results_pair_with_their_own_call_via_call_id(self, dashboard: RailRunDashboard) -> None:
         dashboard.handle_event(
             "tool_call", {"tool": "get_market_data", "arguments": {"symbol": "AAPL"}, "call_id": "call-1"}

agent/tests/test_redaction_cookie_headers.py

@@ -19,6 +19,17 @@ def test_only_the_header_line_is_redacted() -> None:
     assert out == "HTTP/1.1 200 OK\nSet-Cookie: [redacted]\nContent-Type: text/html"
 
 
+@pytest.mark.parametrize("text", [
+    'Cookie: sid="secret"; csrf=another',
+    'Set-Cookie: sid="secret"; Path=/',
+    '''curl -H 'Cookie: sid="secret"; csrf=another' https://example.com''',
+])
+def test_quoted_cookie_value_never_leaks(text):
+    out = redact_text(text)
+    assert "secret" not in out and "another" not in out
+    assert redact_text(out) == out
+
+
 @pytest.mark.parametrize("quote", ['"', "'"])
 def test_quoted_curl_header_keeps_the_rest_of_the_command(quote: str) -> None:
     cmd = f"curl -H {quote}Cookie: sid=abc123; csrf=xyz{quote} https://example.com/api"

agent/tests/test_skill_writer_tools.py

@@ -87,12 +87,13 @@ class TestSaveSkillTool:
         text = (user_dir / "custom" / "SKILL.md").read_text(encoding="utf-8")
         assert "description: Custom skill" in text
 
-    def test_save_forces_frontmatter_name_to_match_slug(self, tool: SaveSkillTool, user_dir: Path) -> None:
+    @pytest.mark.parametrize("field", ["name", "name "])
+    def test_save_forces_frontmatter_name_to_match_slug(self, tool: SaveSkillTool, user_dir: Path, field: str) -> None:
         """A frontmatter name that disagrees with the slug this tool advertises
         would make the skill unreachable under load_skill(slug) and cause
         duplicate entries on every lookup; the saved name must always match.
         """
-        content = "---\nname: MySkill\ndescription: mismatched name\ncategory: user\n---\n\nBody"
+        content = f"---\n{field}: MySkill\ndescription: mismatched name\ncategory: user\n---\n\nBody"
         result = json.loads(tool.execute(name="My Skill", content=content))
         assert result["status"] == "ok"
         text = (user_dir / "my-skill" / "SKILL.md").read_text(encoding="utf-8")

agent/tests/test_trade_journal_fiat_pairs_not_crypto.py

@@ -11,7 +11,7 @@ from src.tools.trade_journal_parsers import _infer_market_from_symbol, parse_gen
 
 @pytest.mark.parametrize(
     "symbol",
-    ["EUR/USD", "GBP/USD", "USD/JPY", "usd-cnh", "AUD-USD", "USD/CHF"],
+    ["EUR/USD", "GBP/USD", "USD/JPY", "usd-cnh", "AUD-USD", "USD/CHF", "USD/BRL", "USD/TWD"],
 )
 def test_fiat_cross_is_not_crypto(symbol: str) -> None:
     assert _infer_market_from_symbol(symbol) == "other"