fix(youtube): address Greptile review feedback

Three changes from automated review on PR #376:

1. Add `--` option terminator before host in _wrap_ytdlp_cmd (P1 security)
   Prevents SSH option injection if LAST30DAYS_YT_SSH_HOST were ever set
   to a value starting with `-` (e.g. `-oProxyCommand=...`). Low
   exploitability since the env var is user-controlled config — but the
   fix is a single arg and turns a self-harm footgun into no footgun.

2. Hoist `import shlex` to module-level (P2 style)
   Pure stdlib import, no reason for the deferred form. Cleaner.

3. Cache _ytdlp_ssh_host() result in fetch_transcript (P2 style)
   Was being called 2-3x per video; the function is cheap (env lookup
   + strip) so this is purely about readability.

Adds test_wrap_cmd_uses_option_terminator covering the security fix
explicitly with a `-oFoo=bar` host value. Updates index assertions in
the two existing tests that check command shape (host is now at index
4, command string at 5, with `--` at 3).

skills/last30days/scripts/lib/youtube_yt.py

@@ -10,6 +10,7 @@ import json
 import math
 import os
 import re
+import shlex
 import shutil
 import sys
 import tempfile
@@ -134,13 +135,14 @@ def _wrap_ytdlp_cmd(cmd: List[str]) -> List[str]:
 
     Args are shell-quoted to survive the remote shell. Uses BatchMode=yes so
     a misconfigured key fails fast instead of hanging on a password prompt.
+    The `--` option terminator prevents an SSH option-injection if
+    LAST30DAYS_YT_SSH_HOST were ever set to a value starting with `-`.
     """
     host = _ytdlp_ssh_host()
     if not host:
         return cmd
-    import shlex
     remote_cmd = " ".join(shlex.quote(a) for a in cmd)
-    return ["ssh", "-o", "BatchMode=yes", host, remote_cmd]
+    return ["ssh", "-o", "BatchMode=yes", "--", host, remote_cmd]
 
 
 def _extract_core_subject(topic: str) -> str:
@@ -520,14 +522,15 @@ def fetch_transcript(video_id: str, temp_dir: str) -> Optional[str]:
     # file on the remote host that we can't easily read back. Skip it and
     # use the HTTP transcript fallback (different YouTube endpoint, less
     # bot-walled, works fine from datacenter IPs).
-    use_ytdlp = is_ytdlp_installed() and not _ytdlp_ssh_host()
+    ssh_host = _ytdlp_ssh_host()
+    use_ytdlp = is_ytdlp_installed() and not ssh_host
     if use_ytdlp:
         raw_vtt = _fetch_transcript_ytdlp(video_id, temp_dir)
         if not raw_vtt:
             _log(f"yt-dlp transcript failed for {video_id}, trying direct HTTP fallback")
             raw_vtt = _fetch_transcript_direct(video_id)
     else:
-        if _ytdlp_ssh_host():
+        if ssh_host:
             _log("SSH-routing active, using direct HTTP transcript fetch")
         else:
             _log("yt-dlp not installed, using direct HTTP transcript fetch")

tests/test_youtube_yt.py

@@ -452,10 +452,13 @@ class TestYtdlpSSHRouting(unittest.TestCase):
         self.assertEqual(wrapped[0], "ssh")
         self.assertEqual(wrapped[1], "-o")
         self.assertEqual(wrapped[2], "BatchMode=yes")
-        self.assertEqual(wrapped[3], "macmini")
+        # `--` terminates SSH option parsing so a host starting with `-`
+        # (e.g. `-oProxyCommand=...`) cannot be reinterpreted as a flag.
+        self.assertEqual(wrapped[3], "--")
+        self.assertEqual(wrapped[4], "macmini")
         # Final arg is the shell-quoted command string
-        self.assertIn("yt-dlp", wrapped[4])
+        self.assertIn("yt-dlp", wrapped[5])
-        self.assertIn("ytsearch5:test", wrapped[4])
+        self.assertIn("ytsearch5:test", wrapped[5])
 
     def test_wrap_cmd_quotes_args_with_spaces(self):
         """Args containing spaces or special chars are shell-quoted."""
@@ -463,7 +466,21 @@ class TestYtdlpSSHRouting(unittest.TestCase):
         cmd = ["yt-dlp", "ytsearch5:hello world", "--dump-json"]
         wrapped = youtube_yt._wrap_ytdlp_cmd(cmd)
         # shlex.quote wraps the whole arg in single quotes when it contains spaces
-        self.assertIn("'ytsearch5:hello world'", wrapped[4])
+        self.assertIn("'ytsearch5:hello world'", wrapped[5])
+
+    def test_wrap_cmd_uses_option_terminator(self):
+        """`--` is inserted before host to prevent SSH option injection.
+
+        Without `--`, an env var like `LAST30DAYS_YT_SSH_HOST=-oProxyCommand=...`
+        would be parsed by ssh as an option flag. The terminator forces it
+        to be treated as a hostname (which will then fail clean if invalid).
+        """
+        os.environ["LAST30DAYS_YT_SSH_HOST"] = "-oFoo=bar"
+        cmd = ["yt-dlp", "--version"]
+        wrapped = youtube_yt._wrap_ytdlp_cmd(cmd)
+        # Find the `--` terminator and verify the host comes immediately after
+        dash_idx = wrapped.index("--")
+        self.assertEqual(wrapped[dash_idx + 1], "-oFoo=bar")
 
     def test_is_ytdlp_installed_short_circuits_with_ssh(self):
         """is_ytdlp_installed returns True without local check when SSH routing is on."""
@@ -489,11 +506,12 @@ class TestYtdlpSSHRouting(unittest.TestCase):
             youtube_yt.search_youtube("test", "2026-02-01", "2026-03-01")
         cmd = run_mock.call_args.args[0]
         self.assertEqual(cmd[0], "ssh")
-        self.assertEqual(cmd[3], "macmini")
+        self.assertEqual(cmd[3], "--")
-        # The shell-quoted yt-dlp invocation lives at index 4
+        self.assertEqual(cmd[4], "macmini")
-        self.assertIn("yt-dlp", cmd[4])
+        # The shell-quoted yt-dlp invocation lives at index 5
-        self.assertIn("--ignore-config", cmd[4])
+        self.assertIn("yt-dlp", cmd[5])
-        self.assertIn("--no-cookies-from-browser", cmd[4])
+        self.assertIn("--ignore-config", cmd[5])
+        self.assertIn("--no-cookies-from-browser", cmd[5])
 
 
 if __name__ == "__main__":