The Agent Skills Directory

[CREDENTIALS_UNSAFE]: The instructions in SKILL.md direct the agent to perform cat aigc-claw/backend/.env | grep -E "API_KEY|KEY". This action reads sensitive API keys directly into the agent's context and conversation logs, increasing the risk of credential exposure.
[DATA_EXFILTRATION]: The Gemini and VLM integration modules (llm_gemini.py and vlm_gemini.py) contain a hardcoded default proxy IP address (35.164.11.19). Routing requests through an unverified third-party intermediary can expose sensitive API keys and user prompts to the proxy operator.
[COMMAND_EXECUTION]: The editor_agent.py script uses subprocess.run to execute ffmpeg for concatenating video segments. While this is a functional requirement, it provides a vector for system-level interaction using external binaries.
[PROMPT_INJECTION]: The skill processes untrusted user-provided story ideas across multiple automated LLM stages. This creates an indirect prompt injection surface where a malicious input could attempt to manipulate the logic or instructions of the subsequent generation agents.

aigc-director