The Agent Skills Directory

Dynamic Module Shimming: The skill performs runtime transformation and loading of JavaScript modules to handle internal dependencies. Specifically, executor-init.ts resolves module paths, rewrites import statements, and uses Base64-encoded data URIs with the import() function to load shims for the @executor-js SDK plugins. This is an architectural mechanism used to resolve packaging constraints in the underlying SDK.
Process Spawning via MCP: The skill supports the Model Context Protocol (MCP) stdio transport, which enables it to spawn and interact with local child processes based on user configuration. This capability allows the agent to interact with local system tools defined in the MCP server configuration.
Remote Resource Ingestion: The skill fetches configuration and schema data from external endpoints, including OpenAPI specification files and GraphQL introspection schemas. These resources are processed to automatically generate tool definitions that the agent can invoke within its environment.
Indirect Prompt Injection Consideration: As a data-bridging tool, the skill processes information from external sources (ingestion points in executor-init.ts and create-executor.ts) that could potentially contain instructions intended to influence the agent. While the skill logic itself does not provide boundary markers, it exposes a significant capability inventory including local process execution and file system access through its tools. To manage these risks, the skill implements a tool approval pipeline (onToolApproval and onElicitation hooks) and uses JSON serialization to ensure that external content is handled as data rather than instructions.

just-bash-executor