The Agent Skills Directory

[CREDENTIALS_UNSAFE]: A hardcoded Coze API key was found in scripts/coze_bot_client.py. Storing credentials in plain text within scripts is a high-risk security practice.\n- [DATA_EXFILTRATION]: The scripts/file_server.py script starts a local HTTP server that listens on all network interfaces (0.0.0.0). This exposes the ./output directory to unauthorized users on the same network.\n- [COMMAND_EXECUTION]: The skill configuration in SKILL.md executes shell commands (pip install -r requirements.txt) during initialization, which can lead to command injection or supply chain attacks if the environment is compromised.\n- [EXTERNAL_DOWNLOADS]: The skill uses scripts/sound_generator.py to fetch audio from api.edata.cloud, a non-whitelisted third-party domain. Downloading media from untrusted sources for further processing (e.g., in moviepy) poses a significant security risk.\n- [REMOTE_CODE_EXECUTION]: The skill attempts path-traversal imports in scripts/voice_generator.py by modifying sys.path to point to a directory outside of the skill's root (../../temp_qwen3_tts/...). This behavior is highly irregular and can be used to execute unintended code.\n- [PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection through its visual analysis workflow.\n
Ingestion points: Video frame extraction in scripts/video_frame_extractor.py followed by analysis in scripts/coze_bot_client.py.\n
Boundary markers: None. Prompts lack instructions to disregard instructions found within processed images.\n
Capability inventory: File system access, network communication, and video composition tools.\n
Sanitization: The skill does not sanitize or validate LLM outputs from the Coze API before further use.

video-recreation