The Agent Skills Directory

[REMOTE_CODE_EXECUTION]: The execute_task tool enables the execution of arbitrary natural language instructions through an autonomous worker with capabilities for coding and DevOps tasks.
[EXTERNAL_DOWNLOADS]: The search_skills tool includes an auto_import parameter that, by default, fetches and stores code from a remote cloud community to the local system.
[DATA_EXFILTRATION]: The upload_skill tool facilitates sending local directory contents to a remote cloud community, with a default visibility of 'public', which may lead to unintended data exposure if misused.
[INDIRECT_PROMPT_INJECTION]: The skill possesses a surface for indirect prompt injection due to its autonomous task processing and skill evolution mechanisms.
Ingestion points: The tool parameters task and query accept untrusted natural language input.
Boundary markers: No delimiters or protective instructions are present to mitigate the execution of malicious commands embedded in task descriptions.
Capability inventory: The delegated agent has extensive permissions for file system access, network communication, and command execution.
Sanitization: No input validation is performed before processing tasks or capturing workflows into reusable skills.

delegate-task