The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: Installs the @membranehq/cli package globally from npm, which is a verified resource provided by the vendor to facilitate the integration.
[COMMAND_EXECUTION]: Utilizes the membrane command-line interface to perform all operations, including login, connection management, and action execution. This centralizes security controls within the vendor's toolset.
[DATA_EXFILTRATION]: While the skill accesses customer interaction data (messages, profiles) from Crisp, it does so through authenticated Membrane connections, and the instructions explicitly advise against requesting or storing user credentials manually.
[INDIRECT_PROMPT_INJECTION]: The skill processes external data from Crisp (such as chat messages and user profiles) which could contain adversarial instructions.
Ingestion points: Crisp messages and profiles are retrieved via list-messages, get-conversation, and list-people-profiles commands.
Boundary markers: No explicit boundary markers or instructions to ignore embedded commands are provided in the skill text.
Capability inventory: The skill can write data back to the external service (e.g., send-message, update-people-profile) and perform arbitrary API requests via the membrane request proxy.
Sanitization: There is no mention of sanitization or escaping of the retrieved external content before it is processed by the agent.

crisp