Automate iterative evaluation and refinement of AI agent outputs with JSON feedback.
npx skills add https://github.com/github/awesome-copilot --skill agentic-eval
Please help me install this Agent Skill. Skill: agentic-eval Source: https://github.com/github/awesome-copilot/tree/main/skills/agentic-eval Command: npx skills add https://github.com/github/awesome-copilot --skill agentic-eval
This Skill enables automated self-evaluation and iterative refinement of AI-generated outputs, reducing errors and enhancing quality through structured critique loops.
Initiate a 3-iteration evaluation cycle on your task: Generate → Evaluate → Critique → Refine → Output
High-intent search queries and answers about installing and using this skill.
Iterative refinement uses self-critique loops where an AI agent evaluates its own code output against quality criteria, identifies issues, and regenerates improved versions. This process repeats over multiple iterations to progressively enhance correctness and reliability before final delivery.
Self-critique is a structured feedback mechanism where an AI evaluates its own work against predefined rubrics, surfaces defects or gaps, and refines the output accordingly. Multiple critique cycles converge toward higher-quality results by catching and correcting errors that single-pass generation misses.
Yes. Automated evaluation applies consistent rubrics to AI-generated artifacts across coding, documentation, and analysis tasks. Structured JSON feedback and configurable iteration limits enable systematic quality checks and refinement loops tailored to your domain requirements.
An evaluator-optimizer workflow separates generation from evaluation: an AI produces initial output, a second evaluator assesses it against criteria, then a refinement loop regenerates improved versions. This separation isolates evaluation logic and enables independent tuning of both stages for higher overall reliability.
Convergence checks detect when successive refinement iterations produce no measurable improvement, signaling that maximum quality has been reached. This prevents wasted iterations and provides a stopping point before hitting your configured iteration limit.
mind-systemsExecute a six-phase reflex-loop workflow to iteratively refine artifacts.
amdCreate export and evaluation plans for Quark quantized models.
BoraPerusicEvaluate LLM outputs with automated metrics and human judgments.
bartolomeomuellerGuide agents through incremental, human-approved changes with stepwise validation.
