The Agent Skills Directory

[DYNAMIC_EXECUTION]: The calculatorTool defined in src/tools.ts utilizes the Function constructor to evaluate mathematical expressions at runtime.
Evidence: const result = Function("\"use strict\"; return (${sanitized})")();
Context: Although a regular expression /[^0-9+\-*\/().\s]/g is applied to sanitize the input, the use of the Function constructor for dynamic code execution is a potentially insecure practice that could be exploited if the sanitization logic is circumvented.
[INDIRECT_PROMPT_INJECTION]: The skill provides an interface for processing user-supplied messages that are subsequently interpreted by the agent and may be passed to the calculatorTool.
Ingestion points: agent.send(input) function within src/agent.ts and triggered in src/headless.ts.
Boundary markers: The skill does not implement explicit delimiters or system instructions to ignore instructions embedded in user data during interpolation.
Capability inventory: Evaluates expressions via the Function constructor in src/tools.ts.
Sanitization: Input is filtered through a whitelist-based regular expression in the tool's execution logic.
[REMOTE_CODE_EXECUTION]: The setup instructions include the installation of third-party libraries from the NPM registry.
Evidence: Installation of @openrouter/sdk, zod, eventemitter3, ink, and react via npm install.
Context: These dependencies are standard libraries consistent with the skill's purpose and are sourced from a well-known package registry.

create-agent