skills-store: Skill for Claude Code

irasutoya-search

Overview

Use the installed irasutoya command-line tool to find Irasutoya illustrations. The skill drives the CLI that is already installed on the system — it does not run, build, or modify any Ruby source. Optimize vague or natural-language requests into short Japanese search candidates before reporting results.

Command Choice

• Use irasutoya random when the user wants an arbitrary Irasutoya image.
• Use irasutoya search "query" when the user gives a topic, keyword, or phrase.
• Use irasutoya help or irasutoya help COMMAND when the user asks what is available.

Expected commands:

irasutoya random
irasutoya search "cat"
irasutoya help search

The search command returns up to 3 matching images. Each result prints:

• Page URL
• Title
• Description
• one or more Image URL lines
• an inline image preview when the terminal supports it

Fast Search Workflow

1. Parse the request into visual facets:
   • subject: person, animal, object, role, age, gender
   • action/pose: observing, wondering, eating, running, pointing
   • emotion/expression: curious, doubtful, happy, surprised, worried
   • props/context: magnifying glass, question mark, school, office
2. Generate 2-4 concise query candidates. Prefer Japanese nouns and short noun phrases because Irasutoya titles/descriptions are Japanese.
3. Search candidates in parallel when possible. Avoid launching many broad searches; stop once a strong match appears.
4. Score results before reporting. Prefer results whose title or description matches at least two facets.
5. Return the best result first, with any close alternative only if it captures a different interpretation.

Pass condition: the first search batch should produce a 4-5 point result in the top two results for at least one candidate query. If it does not, run one refinement batch using the failure rules below.

Query Rubric

Score each candidate result from 0-5:

• +2 subject match: correct character/object type, such as 男の子.
• +1 action/pose match: the visible activity matches, such as 観察.
• +1 emotion/expression match: the description/title implies the requested feeling, such as 不思議そう, 疑問, or はてな.
• +1 prop/context match: important visual prop or setting appears, such as 虫眼鏡.
• +1 single-scene match: one concrete illustration, not a sheet of variations, when the user asked for one image.
• -1 if the result is too generic, multi-image, or only one small sub-image matches.
• -1 if the title includes いろいろな, 表情, or パーツ別 and the user asked for one concrete scene.