The Agent Skills Directory

[DYNAMIC_EXECUTION]: The skill supports executing arbitrary JavaScript snippets via the act --code command. While intended for integration logic and transformations, this provides a surface for the agent to generate and execute dynamic code on the remote platform.
[INDIRECT_PROMPT_INJECTION]: The skill instructs the agent to follow instructions contained in the clientAction.agentInstructions field provided by the external API during connection setup. This creates an attack surface where an external service could influence agent behavior.
Ingestion points: Connection metadata retrieved via npx @membranehq/cli connection get <id> --json (specifically the clientAction object).
Boundary markers: None identified; instructions suggest following the received instructions verbatim.
Capability inventory: The skill has capabilities to perform network requests (act --api), execute scripts (act --code), and modify connection configurations (connection patch).
Sanitization: No sanitization or validation of the external instructions is mentioned.
[COMMAND_EXECUTION]: The skill relies extensively on executing the @membranehq/cli via npx to perform all integration tasks, including authentication, connection management, and action execution.
[EXTERNAL_DOWNLOADS]: The skill fetches the Membrane command-line tool from the npm registry using npx @membranehq/cli or via global installation (npm i -g). These resources are provided by the skill's authoring organization.
[CREDENTIALS_UNSAFE]: The skill documents that authentication credentials for the platform are stored locally in the file path ~/.membrane/credentials.json. It also provides instructions for using a raw Authorization: Bearer token when accessing the API directly.

integrate-anything