Skip to content

"Vibe-Trading: Your Personal Trading Agent"

fix(redaction): tighten URL userinfo and env-name matching

t3148d36b4f9c
diff --git a/agent/src/tools/redaction.py b/agent/src/tools/redaction.py
index 303f5831a..fb2213a91 100644
--- a/agent/src/tools/redaction.py
+++ b/agent/src/tools/redaction.py
@@ -401,9 +401,12 @@ def _redact_token_usage(value: Any, *, sink: str) -> Any:
#: scoped to credential field names so a benign number in shell output (a
#: price, a duration, a timestamp) is never mangled.
#:
-#: All names are SINGULAR and ``\b``-anchored: a plural is a count or a
-#: collection, never a credential, and matching it mangles routine output
-#: (``tokens: 1204 in / 318 out`` from the LLM usage line, ``api_keys: 3``).
+#: All names are SINGULAR and must start a word: either a ``\b`` boundary, or an
+#: UPPERCASE env-style prefix such as ``TUSHARE_`` (see :data:`_TEXT_KV_PATTERN`).
+#: A plural is a count or a collection, never a credential, and matching it
+#: mangles routine output (``tokens: 1204 in / 318 out`` from the LLM usage
+#: line, ``api_keys: 3``). A lowercase suffix like ``pad_token`` or
+#: ``max_token`` is an ordinary identifier, so it is left alone.
_TEXT_CREDENTIAL_KEYS = (
r"api[_-]?key|api[_-]?secret|access[_-]?token|auth[_-]?token"
r"|refresh[_-]?token|bearer[_-]?token|id[_-]?token|session[_-]?token"
@@ -427,7 +430,13 @@ _TEXT_VALUE = r"\"[^\"\n]*\"|'[^'\n]*'|[^\s,;{}\[\]()\"']+"
#: Every quantifier applies to a bounded character class, with no nesting and
#: no overlapping alternatives, so matching stays linear (no ReDoS surface).
_TEXT_KV_PATTERN = re.compile(
- r"(?P<q>[\"']?)(?P<name>" + _TEXT_CREDENTIAL_KEYS + r")\b(?P=q)"
+ r"(?P<q>[\"']?)"
+ # ``\b`` for a plain name; for an env-style name the label follows an
+ # UPPERCASE segment and ``_`` (``TUSHARE_TOKEN``). The inline ``(?-i:...)``
+ # keeps the uppercase test case-sensitive under IGNORECASE, so ``pad_token``
+ # and ``max_token`` still do not match.
+ r"(?:\b|(?<=(?-i:[A-Z0-9])_))"
+ r"(?P<name>" + _TEXT_CREDENTIAL_KEYS + r")\b(?P=q)"
r"(?P<sep>\s*[:=]\s*)"
r"(?![\"']?" + re.escape(_REDACTED) + r"[\"']?|bearer\b)"
r"(?P<value>" + _TEXT_VALUE + r")",
@@ -443,9 +452,13 @@ _TEXT_BEARER_PATTERN = re.compile(
)
#: URL userinfo (``scheme://user:password@host``). It is a credential with no
-#: key label, so the key-based pattern never reaches it. The class stops at the
-#: first ``@`` after ``://``, so an ``@`` later in a path is left alone.
-_TEXT_URL_USERINFO_PATTERN = re.compile(r"(?<=://)[^\s/?#@]+@")
+#: key label, so the key-based pattern never reaches it. The class may contain
+#: ``@`` (a password can), but stops at whitespace, path/query/fragment markers,
+#: quotes, brackets and commas. So the match ends at the last ``@`` of the
+#: authority, and JSON fields after a URL are left alone. A bare username
+#: (``git@host``) is redacted too, since a token used as a username is a
+#: credential.
+_TEXT_URL_USERINFO_PATTERN = re.compile(r"(?<=://)[^\s/?#\"'<>(){}\[\],]+@")
def _sub_credential_pair(match: re.Match[str]) -> str:
@@ -492,9 +505,10 @@ def redact_text(text: object) -> str:
Used for plain-text tool results, shell output, error messages and log
lines where :func:`redact_payload` is a no-op (it only walks structured
keys). Bearer header values and ``key=value`` / ``key: value`` /
- ``"key": "value"`` credential pairs are replaced with ``'[redacted]'``
- while the surrounding text — labels, paths, error context — stays
- readable.
+ ``"key": "value"`` credential pairs are replaced with ``'[redacted]'``,
+ and URL userinfo (``scheme://user:password@host``) becomes
+ ``scheme://[redacted]@host``, while the surrounding text — labels, paths,
+ error context — stays readable.
Args:
text: Free-form string (or any object coerced via :func:`str`).
diff --git a/agent/tests/test_redaction.py b/agent/tests/test_redaction.py
index 99960ad56..9f3ba3cac 100644
--- a/agent/tests/test_redaction.py
+++ b/agent/tests/test_redaction.py
@@ -170,6 +170,16 @@ _URL_USERINFO_SHAPES = (
"GET http://bot:[email protected]/v1 returned 404",
"GET http://[redacted]@hooks.example.com/v1 returned 404",
),
+ (
+ # The password may itself contain ``@``; the match runs to the last one.
+ "see https://user:p@ss@host/x",
+ "see https://[redacted]@host/x",
+ ),
+ (
+ # A token used as the username (no colon) is still a credential.
+ "fetch http://[email protected]/repo",
+ "fetch http://[redacted]@git.example.com/repo",
+ ),
)
@@ -183,7 +193,10 @@ def test_redact_text_scrubs_credential_shapes(raw: str, expected: str) -> None:
assert redact_text(raw) == expected
[email protected]("raw", [shape[0] for shape in _CREDENTIAL_SHAPES])
+ "raw",
+ [shape[0] for shape in _CREDENTIAL_SHAPES + _URL_USERINFO_SHAPES],
+)
def test_redact_text_is_idempotent(raw: str) -> None:
"""Redacting twice must equal redacting once (previews are re-rendered)."""
once = redact_text(raw)
@@ -203,6 +216,10 @@ def test_redact_text_is_idempotent(raw: str) -> None:
"api_keys: 3 configured",
"max_tokens=4096 total_input_tokens: 74812",
"tokenizer=bert-base version=2",
+ # Lowercase suffixes of longer identifiers are not env-style names.
+ "pad_token=0 eos_token=2 max_token=4096",
+ # JSON fields after a URL must survive; only real userinfo is scrubbed.
+ '{"url":"https://api.x.com","email":"[email protected]"}',
],
)
def test_redact_text_leaves_benign_output_readable(raw: str) -> None: