The Agent Skills Directory

[PROMPT_INJECTION]: The skill provides a workflow for scaffolding executable TypeScript plugins based on natural language descriptions from users. This establishes a surface for indirect prompt injection.
Ingestion points: User requests to 'create a plugin', 'write a plugin', or 'make a plugin that does X'.
Boundary markers: The instructions do not provide delimiters or warnings for the agent to ignore potentially malicious instructions within the user's description.
Capability inventory: The skill specifies that generated plugins have access to fetch(), fs, child_process, and process.env.
Sanitization: No sanitization or validation guidelines are provided for handling user-supplied input within the generated execute functions, beyond basic regex in an example.
[REMOTE_CODE_EXECUTION]: The skill documents the platform's ability to load and execute plugins from remote https: URLs and third-party npm packages.
[COMMAND_EXECUTION]: The skill provides templates and instructions for creating plugins that execute shell commands and perform file system operations using Node.js internal modules like node:fs and child_process.

create-plugin