Security audit
Security checks for vulnerabilities and agentic risk
This skill is a straightforward Sarvam AI CLI that sends user-chosen text, audio, and chat prompts to Sarvam's API using the user's API key, with no hidden persistence or destructive behavior found.
Install only if you are comfortable sending selected text, recordings, and chat prompts to Sarvam AI. Do not submit confidential or regulated audio or text unless your use of Sarvam's service is approved, and keep SARVAM_API_KEY scoped and protected.
Credentials or environment variables flow to a network sink. This is a high-confidence indicator of credential exfiltration.
}
try:
response = requests.post(url, json=payload, headers=headers)
response.raise_for_status()
audio_data = response.json()
Credentials or environment variables flow to a network sink. This is a high-confidence indicator of credential exfiltration.
}
try:
response = requests.post(url, json=payload, headers=headers)
response.raise_for_status()
audio_data = response.json()
Credentials or environment variables flow to a network sink. This is a high-confidence indicator of credential exfiltration.
"mode": mode
# "prompt": "" # Optional prompt
}
response = requests.post(url, headers=headers, files=files, data=data)
response.raise_for_status()
print(json.dumps(response.json(), indent=2))
Credentials or environment variables flow to a network sink. This is a high-confidence indicator of credential exfiltration.
}
try:
response = requests.post(url, json=payload, headers=chat_headers)
response.raise_for_status()
result = response.json()
The skill declares executable behavior that uses environment variables and makes external API calls, but it does not declare an explicit tool scope such as permissions or allowed-tools. This creates a transparency and policy-enforcement gap: users and orchestrators may invoke the skill without clear awareness that it can access secrets and transmit data over the network.
The documentation encourages sending text and audio to Sarvam AI for TTS, STT, translation, and chat, but it does not clearly warn users that their inputs are transmitted to a third-party service. This can lead to unintentional disclosure of sensitive or regulated data, especially for audio transcripts, prompts, or API-driven chat content that users may assume is processed locally.
Data is being sent to an external URL. This could be legitimate telemetry or data exfiltration. Manual review is recommended.
}
try:
response = requests.post(url, json=payload, headers=headers)
response.raise_for_status()
audio_data = response.json()
Data is being sent to an external URL. This could be legitimate telemetry or data exfiltration. Manual review is recommended.
}
try:
response = requests.post(url, json=payload, headers=headers)
response.raise_for_status()
audio_data = response.json()
Data is being sent to an external URL. This could be legitimate telemetry or data exfiltration. Manual review is recommended.
}
try:
response = requests.post(url, json=payload, headers=headers)
response.raise_for_status()
audio_data = response.json()
The speech-to-text command reads a local audio file and uploads it to a third-party remote API without any explicit user-facing warning at the point of use. In a CLI handling potentially sensitive recordings, this can lead to accidental disclosure of private or regulated audio content because users may assume processing is local.
Data is being sent to an external URL. This could be legitimate telemetry or data exfiltration. Manual review is recommended.
print(e.response.text)
def chat_completion(message, model="sarvam-2g", system_prompt=None):
url = "https://api.sarvam.ai/chat/completions"
chat_headers = {
"Authorization": f"Bearer {API_KEY}",
The CLI sets default language codes such as hi-IN for text-to-speech and en-IN/hi-IN for translation, which imposes a specific locale behavior when users do not choose one. This is a natural-language policy concern because the file does not offer explicit locale opt-in or explain why India-specific defaults are required.
No suspicious patterns detected.