The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The documentation describes features that ingest natural language "intents" to dynamically generate integration connectors and actions. This identifies a potential surface for indirect prompt injection where crafted input could attempt to influence the agent's generation process.\n
Ingestion points: The intent parameter documented in the connect and action create tools within SKILL.md.\n
Boundary markers: The skill does not explicitly describe boundary markers or "ignore instructions" warnings for the intent processing phase.\n
Capability inventory: The skill documents extensive platform capabilities, including performing network operations, executing JavaScript in a sandbox, and managing workspace settings (SKILL.md).\n
Sanitization: Instructions for sanitizing or validating user-provided intent strings are not included in the skill content.\n- [EXTERNAL_DOWNLOADS]: The skill references and downloads tools and libraries from the vendor's official repositories and well-known registries.\n
Fetches the CLI tool using npx @membranehq/cli.\n
Imports the SDK via the https://esm.sh/@membranehq/sdk CDN.\n
Dependencies include the vendor's packages (@membranehq/sdk, @membranehq/react) and the standard jsonwebtoken library.\n- [COMMAND_EXECUTION]: The skill provides numerous examples of using the vendor's membrane CLI to manage credentials, list connections, and execute actions, which are central to its documented purpose.\n
Commands include membrane credentials, membrane act, and membrane action create.\n
Mentions session storage in ~/.membrane/credentials.json as part of standard CLI authentication flow.

integration-development