The Agent Skills Directory

[DATA_EXFILTRATION]: The script scripts/generate_tts.py is configured to automatically read sensitive API credentials from ~/.alibabacloud/credentials. Combined with the ability to override the base_url parameter via the input request JSON, an attacker could potentially trick the agent into sending the DASHSCOPE_API_KEY to an external, malicious server.
[EXTERNAL_DOWNLOADS]: The skill uses urllib.request.urlopen in scripts/generate_tts.py to download audio files from the URL returned by the DashScope API. While this is expected behavior for a TTS tool, it involves fetching and writing content from a remote source to the local file system.
[INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted user data for audio generation. The request object (containing text, voice, and potentially base_url) serves as an ingestion point. If the agent incorporates instructions from a third party into these parameters without strict validation, it could lead to the exfiltration vector described above or unauthorized file writes.

aliyun-qwen-tts