twalapjs1622/openmeth — reverse-engineered prompt

Build me a working OpenCode plugin that catches the model’s thinking blocks as they stream, like think tags, and reroutes that hidden reasoning through an OpenMythos style recurrent system instead of just letting it pass through normally.

I want it to turn the captured thinking into numpy based representations, run it through a few reasoning stages, including the four archetypes Explorer, Architect, Critic, and Oracle, plus reflection, synthesis, and a recursion step with some kind of convergence check. It should also be able to route reasoning in a few styles, like chain, tree, and graph of thought, then feed useful signals back so the model’s next thinking tokens are guided by that process.

Please keep it lightweight, pure Python, and numpy only, no heavy ML frameworks. Make it easy to configure and install as a normal OpenCode plugin, and include enough tests or a small demo so I can tell the interception and rerouting are actually working. If you need to, look up the current OpenCode plugin docs online.