Testing a codemod-to-agent review handoff without a model

[bomtobe-hgkim]

I saw issue #143 ask how to pass review findings from a run step to the following AI step, and how to test that handoff without a model. A deterministic seam could build the AI prompt from a fixed JSON fixture emitted by the run step, with each finding’s ID, file and line, message, and suggested change. A test can assert that every finding reaches the prompt in stable order without contacting an LLM.
For runs without an agent or API key, a regression test could assert that the review findings still print and the migration keeps its existing exit status. That would cover fallback behavior without model credentials. I haven’t run this workflow; I’m with BOMTOBE, the team building Orke for Claude Code and Codex. This is a test-design suggestion only.