shinka-inspect

Load top Shinka programs and write a Markdown bundle with metadata and code snippets.

1.3k|268|Updated Sep 17, 2025

Search Tags:#planning #markdown #context #agents #ranking #programs #shinka

Ready to run

npx skills add https://github.com/SakanaAI/ShinkaEvolve --skill shinka-inspect

💬 Prompt for ChatGPT / Claude Chat▼

Please help me install this Agent Skill.
Skill: shinka-inspect
Source: https://github.com/SakanaAI/ShinkaEvolve/tree/main/skills/shinka-inspect
Command: npx skills add https://github.com/SakanaAI/ShinkaEvolve --skill shinka-inspect

SYSTEM DOCUMENTATION & REQUIREMENTS

💡 This Skill requires pandas, and includes scripts (resource) components.

What problem does it solve?

Efficiently bring top-performing Shinka programs from a completed run into an agent's working context to accelerate task planning.

Core Features & Use Cases

Extracts top-performing Shinka programs and packages them into a compact Markdown context bundle for downstream agent tasks.
Ranks programs by combined_score and supports optional min_generation filtering, with a fallback to top-k by score if no correct programs exist.
Produces a ready-to-load artifact that agents can consume to guide subsequent mutation planning and evaluation.

Quick Start

Invoke the shinka-inspect skill on a completed run directory to generate a Markdown context bundle for the next mutation planning cycle.

Frequently Asked Questions about shinka-inspect

High-intent search queries and answers about installing and using this skill.

FAQPage Schema

How do I load top-performing Shinka programs from a completed run into agent context?▼

To load top-performing Shinka programs into agent context, read program records from a completed run directory, rank them by combined_score, and select the top-k rows where correct is true to generate a Markdown context bundle.

What is the best way to package ranked program records and code snippets for downstream mutation planning?▼

The best way to package ranked program records for mutation planning is to extract top-k entries by combined_score and produce a Markdown bundle containing metadata, ranking tables, feedback, and code snippets for agents to consume.

How does the fallback ranking work when no correct Shinka programs exist in a completed run?▼

When no correct Shinka programs exist in a completed run, the fallback ranking mechanism selects the top-k programs by combined_score across all available rows, ensuring a Markdown context bundle is still produced for agent planning.

Can I filter Shinka program records by min_generation before ranking them by combined_score?▼

Yes, you can apply optional min_generation filtering to Shinka program records before ranking them by combined_score. This filters the pool of programs prior to selecting the top-k correct entries for the Markdown bundle.

Do I need pandas to extract top-k Shinka programs and write the resulting Markdown bundle?▼

Yes, you need pandas installed to read program records, apply ranking logic by combined_score, and write the resulting Markdown bundle with metadata and code snippets to the specified output path.

What metadata is included in the Markdown bundle generated from top-k Shinka programs?▼

The Markdown bundle generated from top-k Shinka programs includes run metadata, a ranking table of the selected programs, associated feedback, and code snippets, providing comprehensive context for downstream agent tasks.