fix(stt): route transcription inputs through the shared read guard

[Frowtek]

What does this PR do?

transcribe_audio reads a local file and hands it to the configured STT
provider. For the hosted providers (Groq, OpenAI, Mistral, xAI, ElevenLabs)
that ships the file's raw bytes to a third-party API.

The shared local-input read guard (agent.file_safety.get_read_block_error)
was already added to image-gen (587be5b, "guard local provider inputs against
credential reads") and xAI video-gen (1042329), but the STT tools were
missed. So the agent can still feed a .env, auth.json,
.anthropic_oauth.json, or a mcp-tokens/ file to transcription and have its
plaintext leave the machine.

This routes transcribe_audio through the same guard (before validation and
provider dispatch), refusing those paths up front. It is defense-in-depth,
not a security boundary — the guard's own message says so, and the terminal
tool can still reach the file — it just restores parity with the image/video-gen
tools.

Related Issue

N/A — no tracking issue. Parity follow-up to 587be5b (image-gen) and
1042329 (xai video-gen), which added the same guard to sibling tools.

Fixes #

Type of Change

• 🐛 Bug fix (non-breaking change that fixes an issue)
• ✨ New feature (non-breaking change that adds functionality)
• 🔒 Security fix
• 📝 Documentation update
• ✅ Tests (adding or improving test coverage)
• ♻️ Refactor (no behavior change)
• 🎯 New skill (bundled or hub)

Changes Made

• tools/transcription_tools.py — call get_read_block_error(file_path) at the
  top of transcribe_audio, before audio validation / provider dispatch;
  return the guard's error dict when a read is blocked.
• tests/tools/test_transcription_tools.py — add
  TestTranscribeCredentialReadGuard.

How to Test

1. Point the STT path at a secret-bearing file:

   from tools.transcription_tools import transcribe_audio
   r = transcribe_audio("/path/to/.env")
   assert r["success"] is False and "cannot be read" in r["error"]

2. Confirm the block happens before dispatch (the error is the read-guard
   message, not "Unsupported format" and not a provider error).
3. Run the tests:

   pytest tests/tools/test_transcription_tools.py -q
   

Checklist

Code

• I've read the Contributing Guide
• My commit messages follow Conventional Commits (fix(scope):, feat(scope):, etc.)
• I searched for existing PRs to make sure this isn't a duplicate
• My PR contains only changes related to this fix/feature (no unrelated commits)
• I've run the affected suite and it passes: pytest tests/tools/test_transcription_tools.py -q → 107 passed, 1 skipped
• I've added tests for my changes (required for bug fixes, strongly encouraged for features)

Documentation & Housekeeping

• I've updated relevant documentation (README, docs/, docstrings) — N/A (no doc-facing change)
• I've updated cli-config.yaml.example if I added/changed config keys — N/A (no config keys)
• I've updated CONTRIBUTING.md or AGENTS.md if I changed architecture or workflows — N/A
• I've considered cross-platform impact (Windows, macOS) per the compatibility guide — path-agnostic; the guard resolves the same on every OS
• I've updated tool descriptions/schemas if I changed tool behavior — N/A (rejects a narrow set of secret paths; no schema change)

Screenshots / Logs

Mutation check — without the guard, the same .env input instead reaches audio
validation:

- Access denied: .../.env is a secret-bearing environment file and cannot be
  read to prevent credential leakage. ...
+ Unsupported format: . Supported: .aac, .flac, .m4a, .mp3, .mp4, ...

With the guard, transcribe_audio returns the read-guard message before any
provider is contacted.

[alt-glitch]

This was generated by AI during triage.

Duplicate of #57963 — both route local STT input paths in tools/transcription_tools.py through agent.file_safety.get_read_block_error() before cloud provider dispatch (same defense-in-depth parity fix as image-gen/video-gen). #57963 is the earlier open PR (guards inside _validate_audio_file); this one guards at the top of transcribe_audio — same mechanism. Keeping #57963 as canonical; maintainer picks.

[teknium1]

Merged into main via consolidated salvage PR #73514 (merge e807b7106c). Your read-side guard routing transcribe_audio inputs through the shared get_read_block_error was cherry-picked with your authorship (verified gateway temp-dir voice notes pass clean).

Your contribution is credited to you in git history. Thank you! Closing this PR as merged-via-salvage.