Refine AI-generated plans through iterative evaluation and targeted fixes.
npx skills add https://github.com/Tromml/kimchi --skill kimchi-refine
Please help me install this Agent Skill. Skill: kimchi:refine Source: https://github.com/Tromml/kimchi/tree/main/plugins/kimchi/skills/refine Command: npx skills add https://github.com/Tromml/kimchi --skill kimchi-refine
This Skill addresses the challenge of refining AI-generated plans to meet quality thresholds, ensuring they are complete, clear, testable, independent, and appropriately sized before execution.
Use the kimchi:refine skill to improve the plan, allowing up to 5 refinement loops.
High-intent search queries and answers about installing and using this skill.
Refine AI-generated plans using iterative evaluation that scores completeness, clarity, testability, independence, and size. The process automatically applies targeted fixes to the lowest-scoring criteria in a loop until quality thresholds are met.
Evaluating plans for multi-agent execution involves scoring them against five key criteria: completeness, clarity, testability, independence, and size. This ensures tasks are appropriately sized and independent enough for distributed processing.
Automated plan refinement stops based on predefined exit conditions: when a quality score threshold is met, diminishing returns are observed, a regression occurs, or the maximum number of refinement loops is reached.
Iterative refinement is needed after an initial AI plan is generated but before execution. It polishes the plan by identifying weaknesses and applying specific fixes, ensuring the output is robust and actionable for multi-agent execution.
Yes, you can allow up to a specified number of refinement loops, such as 5, to manage the plan quality assurance process. The refinement will automatically exit if this maximum is reached before other exit conditions trigger.
No, automated plan evaluation applies targeted fixes specifically to the lowest-scoring criteria during each iteration. This focused improvement strategy ensures efficient progression toward overall plan quality thresholds.
komlukValidate ResearchPacks and Implementation Plans against predefined quality metrics.
bordenetOrchestrate multi-phase challenges into stress-tested plans with autonomous TODO enrollment.
arittrDecompose feature specifications into phase-based task plans with dependency analysis.
OneWave-AIReview code for security vulnerabilities, performance bottlenecks, and maintainability issues.
