The Agent Skills Directory

[PROMPT_INJECTION]: The skill processes untrusted chat messages from Feishu, which could contain malicious instructions designed to bypass agent constraints (Indirect Prompt Injection).
Ingestion points: The Express server in scripts/feishu-webhook.js receives message events via the /webhook/feishu endpoint.
Boundary markers: The forwarded payload in handleMessageEvent does not wrap user text in protective delimiters or provide the agent with context regarding the untrusted nature of the input.
Capability inventory: According to SKILL.md, the agent is granted access to sensitive tools including Bash, Read, Write, and Edit, which increases the potential impact of a successful injection.
Sanitization: The logic in scripts/feishu-webhook.js extracts the message text and removes @mentions but does not perform content validation, filtering, or sanitization to prevent the execution of embedded commands.

feishu-channel