The Agent Skills Directory

[SAFE]: The skill implements a robust multi-agent review cycle. By requiring consensus between a Planner, Architect, and Critic, the workflow incorporates internal checks and balances that mitigate the risk of a single model producing erroneous or harmful plans.
[SAFE]: A 'Pre-Execution Gate' is defined to intercept underspecified commands. This logic prevents the agent from attempting to execute vague requests (e.g., 'fix the app') without first establishing clear scope and acceptance criteria, which is a key safety measure in autonomous systems.
[SAFE]: The skill manages task context through isolated state files in a local directory (.omx/context/) and delegates actual work to dedicated execution frameworks (Ralph, Team) rather than performing direct, unvetted shell operations.

ralplan