The Agent Skills Directory

[PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection because it ingests and processes untrusted data from external sources (WhatsApp messages).
Ingestion points: Untrusted data enters the context through commands like wacrawl messages, wacrawl search, and wacli messages list, which retrieve content from private and group chats.
Capability inventory: The skill possesses write capabilities through wacli send, which allows for message transmission and state mutation.
Boundary markers: The instructions lack specific boundary markers or delimiters to isolate message content from the agent's primary instructions.
Sanitization: There is no mention of sanitizing or escaping the retrieved message content before it is processed by the agent.
[COMMAND_EXECUTION]: The skill operates by executing local shell commands using the wacrawl and wacli CLI tools.
Command Surface: It defines patterns for searching history, syncing databases, and sending messages.
Data Access: These commands interact directly with the user's private WhatsApp Desktop archives and linked-device session databases located in local project directories (~/Projects/wacrawl and ~/Projects/wacli).

whatsapp