The Agent Skills Directory

[SAFE]: The skill is entirely instructional and aims to improve agent reliability and safety. It defines a '7-step operating reflex' that encourages agents to classify tasks, verify facts before acting, and report evidence for their claims.
[DATA_EXPOSURE]: While the skill mentions using tools like terminal and file access, it does so within a framework of caution (e.g., 'Read the file before editing it', 'Verify a library is actually installed before importing it'). It does not access sensitive credential paths or perform unauthorized network operations.
[PROMPT_INJECTION]: The instructions are designed to prevent agent 'drift' and ensure alignment with the user's original objective. There are no attempts to bypass safety filters or override system constraints; rather, it adds constraints to ensure accuracy.
[COMMAND_EXECUTION]: The skill references standard agent tools (terminal, file, web_search) but does not include any pre-baked malicious commands. It explicitly warns against executing destructive commands like rm -rf without permission.

operating-reflex