The Agent Skills Directory

Dynamic Execution (MEDIUM): The skill performs a dynamic import (await import) using a relative path that traverses three levels above the skill's root directory ('../../../weather/edge'). This reaching outside of the skill's specific directory can be a security concern if the execution environment does not enforce strict isolation between skills or if the directory structure can be influenced by an attacker.\n- Indirect Prompt Injection (LOW): The skill processes and displays data from external market sources which could contain malicious instructions designed to influence the behavior of the AI agent.\n
Ingestion points: Market questions and descriptions retrieved via calculator.scanForEdge() and handled in the scan and top commands within index.ts.\n
Boundary markers: None. Market data is concatenated directly into the markdown output returned to the agent.\n
Capability inventory: No direct subprocess execution, eval/exec calls, or file-write operations were detected in the provided code.\n
Sanitization: The script performs simple string slicing for the output but does not include any logic to escape or validate the content of the external market strings for embedded instructions.

edge