Meta_Kim
A governed execution layer that sits around coding agents such as Claude Code and Codex, turning a plain request into a clarified goal, splitting work across capabilities, reviewing the result, verifying evidence, and writing back lessons for the next run.
Meta_Kim does not replace Claude Code, Codex or other coding agents; it adds a coordination layer around them. Given a request, it first turns it into a goal with success criteria and explicit non-goals, checks which existing agents, skills, commands, tools and project evidence apply, splits independent work when the host supports parallel sessions, then has a review step challenge the implementation and a verification step check fresh evidence before anything is called done, keeping unproven claims visible rather than hidden.
Useful lessons from a run can be written back through a governed approval path, so a later run starts with stronger capabilities instead of no memory of what happened before. A local, read-only control room shows one execution graph with named sessions, a dependency graph of the current run, a four-column work board, and a project or session repository view, along with tool activity, blockers, outputs, evidence and replay history.