The Agent Skills Directory

[COMMAND_EXECUTION]: The skill defines a daemon mode and an autonomous loop for persistent background execution. The logic involves scanning the host environment and executing actions which suggest shell command usage for builds and tests.
[REMOTE_CODE_EXECUTION]: The skill implements a self-modifying agent that generates code improvements and executes them dynamically on its own source code. This dynamic execution of AI-generated content is a significant security risk.
[PROMPT_INJECTION]: The autonomous loop creates a surface for indirect prompt injection where data scanned from the environment can influence future autonomous actions. Ingestion points: File system and environment state scanned in rules/autonomous-loop.md. Boundary markers: None present in the provided instructions. Capability inventory: The execute method performs file writes and system status checks based on generated decisions. Sanitization: No sanitization or verification of the generated code or environment data is present.

zi-kong