hiddink-harness:auto-improve

Automating agent configuration improvements involves parsing eval-core analysis data to identify changes, applying them in isolated git worktrees, and enforcing verification before generating a pull request. This skill executes that entire workflow automatically.

Isolated git worktree verification works by applying routing logic updates to a separate working directory, running mandatory mgr-sauron pipeline checks, and only generating a PR after validation succeeds. This prevents unverified changes from affecting the main branch.

Yes, you need both git and eval-core CLI integration to automate agent optimization. Git manages state transitions and isolated worktrees for safe changes, while eval-core CLI provides the analysis data required to identify and select pending configuration improvements.

The best way to apply model escalation for failed benchmarks is to automatically parse evaluation results, apply the recommended model or routing update to an isolated worktree, verify the fix through the Sauron pipeline, and open a PR for final review.

No, you cannot use this automated configuration optimization without git-based version control. The skill requires git to manage state transitions, create isolated worktrees for safe execution, and generate pull requests for final human review.