The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted data from external sources (such as tweets, bios, and messages) through the execute tool.
Ingestion points: Data returned by the execute tool is incorporated into the agent's context as described in SKILL.md.
Boundary markers: The instructions require the agent to quote and label external content and explicitly ignore embedded instructions.
Capability inventory: The execute tool allows for network requests and state-changing operations via xquik.request.
Sanitization: The skill mandates that external text must not influence tool selection, targets, or credentials.
[DYNAMIC_EXECUTION]: The search and execute tools use async arrow functions passed as strings ('code') to define logic for API requests and schema discovery.
The execute tool takes an arrow function that calls xquik.request({ path, method, query, body }).
This dynamic construction is limited to the predefined xquik object methods and is documented as the primary method for interacting with the remote service.

xquik-mcp