The Agent Skills Directory

[DATA_EXFILTRATION]: The instructions identify the specific filesystem directory (~/.openclaw/openclaw-weixin/accounts/) used for storing sensitive WeChat authentication credentials. Disclosing these credential locations to an AI agent presents a data exposure risk if the environment allows for unauthorized file access.
[PROMPT_INJECTION]: The skill architecture is vulnerable to indirect prompt injection through external IM channels. 1. Ingestion points: Untrusted user data enters the system context via Telegram, Discord, WeChat, and other integrated messaging platforms (SKILL.md). 2. Boundary markers: The skill does not provide the agent with instructions or delimiters to isolate content received from these channels or to ignore embedded instructions. 3. Capability inventory: The agent is authorized to use the config.patch tool, which can modify global system configurations, and authentication-related tools like weixin_login. 4. Sanitization: No sanitization or validation processes are described for content received from external IM channels before it is used to influence the agent's logic or configuration changes.

Channels Guide