The Agent Skills Directory

[REMOTE_CODE_EXECUTION]: The skill incorporates a specialized sub-skill (execute-code-sync.md) and a support script (scripts/execute_code_sync_request_example.py) designed to transmit and execute arbitrary code on a remote server via a POST request to an API endpoint. This process involves the dynamic generation and execution of Python, JavaScript, and Shell scripts to perform complex data processing.
[REMOTE_CODE_EXECUTION]: The provided Python script execute_code_sync_request_example.py includes an --insecure flag that disables SSL certificate verification and hostname checking (ssl.CERT_NONE), which can expose the communication to man-in-the-middle attacks.
[COMMAND_EXECUTION]: The smart-ask-data.md and smart-search-tables.md sub-skills extensively use the kweaver CLI tool to manage knowledge base configurations, search for metadata, and execute database queries.
[PROMPT_INJECTION]: The skill presents an attack surface for indirect prompt injection by processing untrusted user queries and interpolating them into SQL statements and code execution payloads.
Ingestion points: User input entering the system via the smart-data-analysis entry point and flowing into sub-skills (e.g., in smart-ask-data.md).
Boundary markers: No explicit delimiters or boundary markers are defined to separate user-provided strings from the generated logic.
Capability inventory: The agent can execute database queries and run remote code, providing high-impact capabilities that can be abused via malicious input.
Sanitization: The instructions lack specific requirements for sanitizing or validating user input before its use in command or query construction.
[DATA_EXFILTRATION]: The skill is designed to move data between the user's environment, internal databases, and remote processing APIs. The execute_code_sync_request_example.py script facilitates the transmission of potentially sensitive session information, authentication tokens, and result data to remote infrastructure.

smart-data-analysis