The Agent Skills Directory

[PROMPT_INJECTION]: The skill contains highly contradictory metadata. The YAML frontmatter claims the skill is for managing 'Persons, Organizations, Deals, Leads, Projects, Activities', which are CRM-related entities. However, the body of the skill describes 'Poof' as a 'disappearing message app' similar to Snapchat. This deceptive documentation can lead to user confusion regarding the types of data the skill interacts with.
[PROMPT_INJECTION]: The skill is vulnerable to indirect prompt injection. It ingests untrusted data from external services via the Membrane CLI.
Ingestion points: Data enters the agent context via the output of membrane action run and membrane action list commands.
Boundary markers: None. There are no instructions to the agent to treat the output of these tools as untrusted or to use delimiters.
Capability inventory: The skill can execute arbitrary actions via membrane action run, create new actions via membrane action create, and perform network-based logins.
Sanitization: None. The skill does not specify any filtering or validation for the data returned by the external actions before it is processed by the agent.
[COMMAND_EXECUTION]: The skill relies on the execution of the membrane CLI tool for all its operations, including authentication (membrane login), connection management (membrane connect), and action execution (membrane action run).
[EXTERNAL_DOWNLOADS]: The instructions direct the user to install a global Node.js package (@membranehq/cli@latest) from the npm registry to enable the skill's functionality.

poof