.Agent

Recombine proven parts of multiple evaluated attempts into a coherent follow-up attempt. Use when ClimbHill attempt comparison reveals complementary partial wins.

Analyze ClimbHill runs, attempts, lineage, costs, failures, and recurring constraints. Use for experiment-history diagnosis and evidence-backed follow-up proposals.

Ground implementation decisions in recorded ClimbHill resources and cited prior work. Use when code changes should trace back to research, guidance, postmortems, or experiment evidence.

Classify proposed paths and enforce denied or approval-required ClimbHill policy before edits. Use when work may touch tests, CI, infrastructure, secrets, or other protected surfaces.

Coordinate independent, bounded coding attempts against one goal and compare them after policy and evaluation checks. Use for parallel ClimbHill improvement attempts.

Produce evidence-backed run and attempt reports with evaluations, costs, risks, and promotion recommendations. Use at the end of a ClimbHill run or attempt attempt.

Sample successful and failed historical ClimbHill Attempts before planning new work. Use when prior Runs may contain reusable approaches or known failure modes.

Review policy, evaluations, cost, lineage, and risk before recording a human promotion or rejection decision. Use when a ClimbHill attempt is ready for disposition.

Convert evidence-backed ClimbHill experiment findings into reviewable GitHub issue proposals. Use after experiment-tree analysis identifies actionable repository improvements.

Prepare a repository for controlled recursive optimization by defining goals, agent guidance, policies, evaluations, resources, and protected surfaces. Use when a codebase lacks ClimbHill support files.