gsd-plan-review-convergence

Automating AI code review for iterative planning requires a convergence loop that repeatedly executes plan reviews and re-invokes the planning phase using review feedback until no HIGH concerns remain or a maximum cycle limit is reached.

A replanning loop for risk triage continuously evaluates project plans through external reviewer CLIs, detecting HIGH concerns and feeding the review feedback back into the planning phase to converge on an acceptable risk state.

Yes, you can use local OpenAI-compatible model servers like Ollama, LM Studio, or llama.cpp alongside selectable AI CLI reviewers such as Codex, Gemini, or Claude to execute the cross-AI external review loop.

You configure workflow gating limits by specifying a phase number and setting the max-cycles parameter, which enforces revision-loop behavior, stall detection, and escalation boundaries to stop the convergence loop safely.

Automated planning loops stall during risk triage when the workflow gating safeguards detect a lack of convergence progress or when the maximum cycle limit is reached before eliminating all HIGH concerns from the plan.

Use an external review-driven replanning loop when you need to enforce strict quality-gating workflows across development cycles, ensuring that automated cross-AI review feedback systematically eliminates HIGH risk concerns that manual checks might miss.