optimize-skill

To fix incorrect tool selection in AI agents, use an eval-driven optimization methodology to audit the skill instructions, run comparative baseline evaluations, and iterate until the agent consistently chooses the correct tool.

Eval-driven optimization improves AI agent skills by systematically measuring baseline performance, testing against defined cases, and refining instruction routing and conciseness to ensure measurable performance gains.

Benchmark agent workflow performance by running parallel subagents to measure current skill metrics, then applying an audit-baseline-fix-verify loop to guarantee a 100% pass rate against defined test cases.

Optimizing routing logic requires defined test cases for systematic baseline measurement and verification, ensuring that any improvements to complex agent workflows are verified against baseline metrics.

The best way to reduce instruction verbosity is an automated verification loop that targets instruction conciseness, auditing existing skills and fixing routing logic until measurable performance gains are achieved.

Agent performance degrades with stale instructions due to poor routing logic and verbosity, requiring an eval-driven methodology to audit, fix, and verify skills against baseline metrics.