Evaluate and refine AI agent outputs through iterative self-critique loops.
npx skills add https://github.com/eric-sabe/engsys --skill agentic-eval-eric-sabe
Please help me install this Agent Skill. Skill: agentic-eval Source: https://github.com/eric-sabe/engsys/tree/main/core/skills/agentic-eval Command: npx skills add https://github.com/eric-sabe/engsys --skill agentic-eval-eric-sabe
This Skill solves the problem of inconsistent, low-quality AI agent outputs for quality-critical tasks by providing structured, repeatable patterns for evaluating and refining generated content until it meets predefined standards.
Use the agentic-eval skill to build a self-critique loop that refines your generated code until it passes all predefined quality checks.
High-intent search queries and answers about installing and using this skill.
You can automatically refine AI-generated code by implementing iterative self-critique loops that evaluate outputs against predefined test cases and refine them through multiple passes until all tests pass. This approach uses evaluator-optimizer pipelines to separate generation from quality control.
The best way to set up an evaluator-optimizer pipeline for report drafting is to separate generation and evaluation responsibilities, applying weighted rubric-based assessment to automatically refine the draft. This ensures clear, maintainable quality control through iterative refinement loops.
Yes, you can use self-critique loops for any quality-critical task requiring adherence to defined success criteria, including report drafting and data analysis. The flexible evaluation strategies support outcome-based scoring and LLM-as-judge comparison to match different task types.
No specific dependencies are required to implement rubric-based scoring in agent workflows. The Skill provides structured, repeatable patterns for evaluating generated content directly, allowing you to apply weighted rubric-based assessment without external modules.
Iterative refinement may stop before meeting quality standards if the evaluation criteria are poorly defined or the generation lacks sufficient context. Ensure your rubric-based scoring and evaluator-optimizer pipelines have clear, measurable success criteria to avoid premature termination.
m2ai-portfolioGenerate evaluation criteria and rubrics for knowledge-work outputs.
FriendlyInternetMap Nuxt Crouton GitHub workflows and Cloudflare Workers deploy pipeline.
oren-arguBuild Trigger.dev agent workflows with orchestration, routing, and human-in-the-loop gates.
xabierlameiroEvaluate and refine AI agent outputs through self-critique and iterative loops.
