Where is the optimization progress stored?

All optimization state, including hypotheses and attribution confidence, is persisted in a centralized journal.yaml file within the agent's specific configuration directory.

How does this skill prevent performance regressions?

It enforces a strict decision policy where any violation of guardrails or regression in core metrics triggers an automatic rollback to the last known good state.

What is the difference between single and multi-lever mode?

Single mode restricts hypotheses to one variable for isolated testing, while multi-lever mode allows for tuning 2 to 5 variables simultaneously to find complex optimizations.

What is the purpose of the preflight check?

The preflight check validates your agent's evaluation contract and live identifiers before optimization begins, ensuring the environment is correctly configured to prevent failed runs.

Optimization Craft

Name: Optimization Craft
Author: mberto10

bymberto10

0•

데이터 과학 및 ML

Refines AI agent performance through structured optimization loops, automated evaluation contracts, and persistent multi-lever tuning.

Optimization Craft provides a rigorous framework for the /optimize command, ensuring that agentic improvements are data-driven and verifiable. It enforces a mandatory preflight phase to validate evaluation contracts before execution and manages the optimization process through five distinct phases: Initialize, Hypothesize, Experiment, Analyze, and Compound. By supporting both single and multi-lever tuning strategies and maintaining a persistent state journal, it prevents regressions while allowing for complex, multi-dimensional performance gains in AI workflows.

주요 기능

01Configurable single and multi-lever tuning strategies for up to 5 variables

02Standardized iterative loop flow for hypothesis-driven agent improvement

03Mandatory preflight contract validation to ensure evaluation readiness

04Strict regression guard policy with automatic rollback capabilities

05Automated state journaling in YAML for persistent progress tracking

060 GitHub stars

사용 사례

01Tuning AI agent prompts and parameters based on specific evaluation datasets

02Ensuring consistent performance tracking across long-running optimization cycles

03Conducting multi-dimensional experiments to optimize agent success rates

What are Skills?·How to Install

Install with 🐟 Skill.Fish

npx skillfish add mberto10/mberto-compound optimization-craft

For use in Claude.ai and ChatGPT

Download Skill