dsh-llm-verifier

In modern development, relying on a single coding agent to generate patches can be limiting—it may miss simpler fixes, better tests, or subtle regressions. Manually creating and comparing multiple candidates is tedious and often inconsistent. The dsh-llm-verifier plugin addresses this by automating the entire process into an approval-gated workflow. It runs 3 or 5 independent coding candidates in isolated Git worktrees, validates each against project tests to discard failures, and then uses an LLM verifier to rank the remaining patches intelligently. The outcome is an auditable report that details rankings, changed files, and token usage, ensuring transparency. Users maintain control with two explicit approval gates: one before candidate execution and another before applying the winner, along with built-in security features like credential isolation and integrity checks. This plugin is ideal for developers or teams tackling high-value coding tasks—such as critical features or complex bug fixes—who seek more reliable patch selection. It is particularly suited for trusted repositories, as validation commands execute code on the host, emphasizing the need for caution and security awareness.