The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill instructs the user to install and run the official Membrane CLI package from the NPM registry.
Evidence: npm install -g @membranehq/cli@latest and npx @membranehq/cli@latest. These resources are owned by the vendor 'membranedev' and are required for the skill's operation.
[COMMAND_EXECUTION]: The skill relies on the execution of the membrane CLI tool to perform all operations, including authentication, connection management, and running actions.
Evidence: Multiple command examples throughout SKILL.md demonstrate the use of subprocess calls to the membrane binary to interact with external data.
[REMOTE_CODE_EXECUTION]: The skill leverages the Membrane platform's ability to dynamically generate and execute logic based on natural language descriptions.
Evidence: The membrane action create "DESCRIPTION" command allows for the runtime generation of new actions. This behavior is a core feature of the Membrane platform and is handled via the vendor's infrastructure.
[PROMPT_INJECTION]: The skill possesses an indirect prompt injection surface where untrusted data is processed.
Ingestion points: User-supplied QUERY strings in membrane action list and DESCRIPTION text in membrane action create are interpolated directly into shell commands within SKILL.md.
Boundary markers: No explicit delimiters or instructions to ignore embedded commands are present in the command templates.
Capability inventory: The skill can execute shell commands, perform network requests through connections, and trigger remote actions.
Sanitization: There is no evidence of sanitization or validation of the input strings before interpolation.

dingconnect