From 050461ec83384c700507c59b01c8517edd607e6e Mon Sep 17 00:00:00 2001
From: Teknium <127238744+teknium1@users.noreply.github.com>
Date: Tue, 28 Jul 2026 12:01:53 -0700
Subject: [PATCH] fix: hoist STT credential read guard to the public
 transcribe_audio entry point

The rebase onto #73510's prepare/dispatch split left the guard inside
_transcribe_prepared_audio, where source validation ran first and a
blocked .env surfaced a format error instead of the read-block message.
Guard now fires before any validation/preprocessing.
---
 tools/transcription_tools.py | 9 +++++++++
 1 file changed, 9 insertions(+)

diff --git a/tools/transcription_tools.py b/tools/transcription_tools.py
index 2dc88f78fbb..54acdf72f7b 100644
--- a/tools/transcription_tools.py
+++ b/tools/transcription_tools.py
@@ -2397,6 +2397,15 @@ def _transcribe_prepared_audio(file_path: str, model: Optional[str] = None) -> D
 
 def transcribe_audio(file_path: str, model: Optional[str] = None) -> Dict[str, Any]:
     """Safely validate, preprocess supported inputs, and dispatch transcription."""
+    # Refuse to feed a credential / secret store (auth.json, .env, OAuth
+    # tokens, mcp-tokens/, ...) to an STT provider — before ANY validation or
+    # preprocessing, so the refusal names the real reason rather than a
+    # format error. Mirrors the image-gen / video-gen read guards.
+    from agent.file_safety import get_read_block_error
+    blocked = get_read_block_error(file_path)
+    if blocked:
+        return {"success": False, "transcript": "", "error": blocked}
+
     # Cap .silk sources before the decoder runs (decoder safety). For all
     # other inputs the remote-upload size cap is provider-scoped and enforced
     # in _transcribe_prepared_audio, so local whisper can handle big files.