end-of-turn-review

Automating code review for a git diff requires a hook-based script that evaluates newly produced changes for correctness and edge cases. This process applies end-of-turn review passes to your working tree, returning structured verdicts with specific issues and file references before you finalize a commit.

Plan critique is the evaluation of proposed implementation plans for hidden dependencies, missing failure handling, and load-bearing mistakes. You need this second-opinion review pass before executing a plan to catch weak assumptions, bugs, and regressions early in the design phase.

Yes, you need LiteLLM to access the reviewer model for automated LLM reviewer scripts. The workflow requires a git working tree, shell-based hook execution via bash, and LiteLLM connectivity to generate structured verdict outputs from the code diff analysis.

Yes, you can use bash scripts and git hooks for pre-commit code review checks. This approach fits end-of-turn automation by running shell-based scripts over your working tree to catch regressions and unsafe assumptions without directly editing the work itself.

The best way to catch regressions and broken assumptions is applying a second-opinion review pass over newly produced work. By analyzing non-trivial code changes through a reviewer model, you get a structured verdict highlighting specific issues and file references before they reach a commit.

Limitations of using shell scripts for diff analysis include dependencies on a git working tree and external tools like grep, jq, and curl. The process requires shell-based hook execution access and cannot review changes outside the current working tree or without LiteLLM model connectivity.