The Agent Skills Directory

[PROMPT_INJECTION]: The skill incorporates an 'Untrusted data guard' that wraps trace-derived text and codebase content in <trace-context> delimiters. This design pattern is a security best practice intended to prevent indirect prompt injection by ensuring the agent treats processed data as context rather than instructions.
[COMMAND_EXECUTION]: Orchestration is performed by invoking other platform skills (such as oh-my-claudecode:trace and oh-my-claudecode:omc-plan) via the standard Skill() function. This is a controlled execution of existing tools within the platform's ecosystem.
[DATA_EXFILTRATION]: The skill reads local project information and writes state/specifications to the .omc/ directory. No unauthorized network operations or data transmission to external domains were identified.

deep-dive