Two inherited defects: --fingerprint always 400'd, and the shipped credential check was dead

.github/ci_checks.py

@@ -53,7 +53,13 @@
 
 # Documented placeholders, which are supposed to look like the real thing.
 CREDENTIAL_ALLOWED = ("USER:PASS", "user:pass", "ACCOUNT:PASSWORD", "{login}",
-                      "***", "password}@", "u:p@h", "LOGIN:PASSWORD")
+                      "***", "password}@", "u:p@h", "LOGIN:PASSWORD",
+                      # These were missing, which is why `--all` failed on
+                      # this repo's own main: this file matches `http://` as
+                      # well as `ws://`, and the documentation and masking
+                      # fixtures use `http://` placeholders. A check that
+                      # fails on its own repository is one nobody can read.
+                      "{user}", "user:secret@", "login:password@host:port", "myuser:s3cr3t@", "u:supersecret@", "login:supersecret@", "u:pass@h1", "u:pass@h2")
 
 # A 2captcha API key is a 32-character hex string.
 HEX32 = re.compile(r"\b[0-9a-f]{32}\b")

.github/workflows/tests.yml

@@ -132,20 +132,16 @@ jobs:
       - name: No credentials committed
         # A key pasted into a file during debugging is the one mistake in this
         # repo that cannot be undone by a later commit.
-        run: |
-          set -e
-          if grep -rInE '(ws|wss)://[^ "'"'"']+:[^ "'"'"']+@' \
-               --include='*.py' --include='*.md' --include='*.txt' --include='*.yml' . \
-             | grep -vE 'USER:PASS|user:pass|ACCOUNT:PASSWORD|\{login\}|\*\*\*|password\}@'; then
-            echo "FAILED: what looks like a real credentialled URL is committed"
-            exit 1
-          fi
-          if grep -rIn --include='*.py' --include='*.md' -E '\b[0-9a-f]{32}\b' . \
-             | grep -viE 'sha|hash|nonce|example|md5'; then
-            echo "FAILED: what looks like a 2captcha API key is committed"
-            exit 1
-          fi
-          echo "clean"
+        #
+        # THIS RUNS THE SHIPPED CHECK, not a copy of it. There used to be an
+        # inline grep here with its own allowlist, and the two disagreed in a
+        # way that mattered: the inline version matched only `ws://` and
+        # `wss://`, so an `http://user:pass@` credential would have sailed
+        # past CI, while `.github/ci_checks.py` — which does match `http://`
+        # — was in the repo and invoked by nothing at all. Two sources of
+        # truth, one of them dead, and the live one narrower than the dead
+        # one.
+        run: python3 .github/ci_checks.py --secret-check
 
   engine-smoke:
     # The offline job above installs NO engine library on purpose (that's

playwright_scraper.py

@@ -949,8 +949,21 @@ def parse_args():
                    help="Fetch a browser fingerprint from 2captcha's Fingerprint API and "
                         "apply it to the launched browser. Needs --twocaptcha-key. Ignored "
                         "with --cdp-endpoint, where the Scraping Browser supplies its own.")
-    p.add_argument("--fp-tags", default="Windows,Chrome,Desktop",
-                   help="Fingerprint filter tags (default: Windows,Chrome,Desktop)")
+    # ONE OS-family tag, not a list — and this default is what makes
+    # --fingerprint work at all. It shipped as "Windows,Chrome,Desktop",
+    # which the fingerprint API rejects with HTTP 400 ("Request parameters
+    # are invalid"), so --fingerprint failed on every invocation.
+    #
+    # fingerprint_client.py's own --tags help has said so all along; the
+    # engines' default contradicted it. Measured against the live API on
+    # 2026-09-10: `Windows` succeeds, and `Windows,Chrome,Desktop`,
+    # `Chrome` and `Desktop` each 400.
+    p.add_argument("--fp-tags", default="Windows",
+                   help="ONE OS-family tag for the fingerprint filter: "
+                        "Windows, Microsoft Windows or Android. NOT a list — "
+                        "Chrome, Desktop and Mobile are each rejected by the "
+                        "API with 400, and no combination is accepted. Use "
+                        "--fp-country to narrow further. (default: Windows)")
     p.add_argument("--fp-country", default=None,
                    help="Fingerprint country, ISO 3166-1 alpha-2. Match it to your proxy's "
                         "exit country — a US fingerprint on a German IP is a contradiction.")

smoke_test.py

@@ -18,6 +18,7 @@
 import re
 import builtins
 import inspect
+import subprocess
 import sys
 import tempfile
 
@@ -2070,6 +2071,57 @@ class _NoProxyArgs:
     ok &= check("the placeholder antidetect endpoint constant is gone",
                 not hasattr(_cs, "ANTIDETECT_LOCAL_API"))
 
+    # THE SHIPPED CI CHECKS RUN, AND PASS ON THIS REPO.
+    #
+    # `.github/ci_checks.py` was in this repo and invoked by NOTHING — not
+    # CI, not this suite — while `tests.yml` carried an inline grep doing a
+    # narrower version of the same job with its own allowlist. The inline one
+    # matched only ws:// and wss://, so an `http://user:pass@` credential
+    # would have sailed past CI; the shipped one, which does match http://,
+    # meanwhile failed on this repo's own main because the documentation
+    # placeholder in proxy_pool.py and the masking fixtures here were missing
+    # from its allowlist.
+    #
+    # Two sources of truth, one dead and one with a hole. Running the shipped
+    # one here as well means a failure shows up locally, before a push.
+    _repo_root = os.path.dirname(os.path.abspath(__file__))
+    _script = os.path.join(_repo_root, ".github", "ci_checks.py")
+    ok &= check("ci_checks.py is present", os.path.exists(_script))
+    if os.path.exists(_script):
+        _proc = subprocess.run([sys.executable, _script, "--all"],
+                               cwd=_repo_root, capture_output=True, text=True)
+        ok &= check(f"ci_checks.py --all passes on this repo "
+                    f"(exit {_proc.returncode})", _proc.returncode == 0)
+        if _proc.returncode != 0:
+            for _line in (_proc.stdout + _proc.stderr).strip().split("\n")[-12:]:
+                print(f"        {_line}")
+        _wf = open(os.path.join(_repo_root, ".github", "workflows",
+                                "tests.yml"), encoding="utf-8").read()
+        ok &= check("tests.yml runs the shipped check rather than an inline "
+                    "copy",
+                    "ci_checks.py --secret-check" in _wf
+                    or "ci_checks.py --all" in _wf)
+        ok &= check("...and carries no second, narrower inline credential "
+                    "grep", "(ws|wss)://[^ " not in _wf)
+
+    # `--fp-tags` MUST DEFAULT TO ONE OS-FAMILY TAG. It shipped as
+    # "Windows,Chrome,Desktop", which the fingerprint API rejects with HTTP
+    # 400 — so --fingerprint failed on every invocation, while
+    # fingerprint_client.py's own --tags help said ONE tag all along.
+    # Measured against the live API on 2026-09-10: `Windows` succeeds, and
+    # `Windows,Chrome,Desktop`, `Chrome` and `Desktop` each 400.
+    import glob as _glob
+    for _path in sorted(_glob.glob(os.path.join(_repo_root, "*_scraper.py"))):
+        _m = re.search(r'--fp-tags"\s*,\s*default="([^"]*)"',
+                       open(_path, encoding="utf-8").read())
+        if _m is None:
+            continue
+        ok &= check(f"{os.path.basename(_path)}'s --fp-tags default is ONE "
+                    f"tag the API accepts",
+                    "," not in _m.group(1)
+                    and _m.group(1) in ("Windows", "Microsoft Windows",
+                                        "Android"))
+
     print()
     if _skips:
         print(f"{len(_skips)} group(s) of checks SKIPPED — an optional engine "