sk-improve-agent

What problem does it solve?

This skill enables an evaluator-first loop to improve bounded agent surfaces safely, preventing direct canonical edits and ensuring evidence-backed changes.

Core Features & Use Cases

• 5-dimension scoring across structural integrity, rule coherence, integration consistency, output quality, and system fitness to quantify improvements.
• Dynamic target profiling that derives evaluation rules from the agent's own frontmatter and policy, enabling on-the-fly scoring for any agent in the repository.
• Guarded promotion and rollback with append-only evidence, traceable benchmarks, and memory artifacts that support auditable decisions.
• Integration-scanning and drift visibility to keep runtime mirrors aligned while preserving evaluation truth.

Quick Start

Run a complete improve-agent loop against a target agent path to produce a packet-local candidate and associated evidence before mutating the canonical target.