Gemini STT

AdvisoryAudited by Static analysis on Apr 30, 2026.

Overview

No suspicious patterns detected.

Findings (0)

Artifact-based informational review of SKILL.md, metadata, install specs, static scan signals, and capability signals. ClawScan does not execute the skill or run runtime probes.

Concern

ASI02: Tool Misuse and Exploitation

What this means

If the agent is tricked into using an unsafe Vertex region value, a Google Cloud access token and the audio being transcribed could be sent to a non-Google host.

Why it was flagged

The user-controlled region value is inserted into the URL host while a Google bearer token is attached. A malicious or manipulated region containing a slash could change the request destination and receive the token and audio payload.

Skill content

url = f"https://{region}-aiplatform.googleapis.com/v1/projects/{project}/locations/{region}/publishers/google/models/{model}:generateContent"
...
"Authorization": f"Bearer {access_token}",

Recommendation

Validate region against an allowlist or strict region-name regex before constructing the URL, and refuse values containing URL metacharacters such as '/', ':', '@', '?', or '#'. Require confirmation for non-default endpoints.

Note

ASI03: Identity and Privilege Abuse

What this means

The skill may use your Google API key or active Google Cloud account/project, which can affect billing and access permissions.

Why it was flagged

The skill uses either a Gemini API key from the environment or a local gcloud access token. This is expected for the stated Google transcription integration, but it is sensitive credential use.

Skill content

api_key = os.environ.get("GEMINI_API_KEY")
...
["gcloud", "auth", "print-access-token"]

Recommendation

Use a least-privileged API key or Google Cloud project, verify the active gcloud account and project before running, and declare these credential requirements in the skill metadata.

Note

ASI07: Insecure Inter-Agent Communication

What this means

Audio contents you transcribe are sent to Google for processing.

Why it was flagged

The selected local audio file is read, base64-encoded, and included in a request to Google's Gemini or Vertex AI API. This is purpose-aligned but crosses a privacy boundary.

Skill content

with open(file_path, "rb") as f:
    audio_data = f.read()
...
{"inline_data": {"mime_type": mime_type, "data": b64_data}}

Recommendation

Only use this skill for audio you are comfortable sending to Google, and review Google's Gemini or Vertex AI data handling terms for your account type.