irasutoya-search
Skill13 starsChanged 3 months agoMITView on GitHub
Search and preview Irasutoya illustrations with the installed `irasutoya` command-line tool. Use this skill whenever the user wants to find, recommend, preview, randomly pull up, or get the page/image URL of an Irasutoya (いらすとや) image — from a topic, scene, Korean/Japanese/English phrase, visual description, emotion, character, object, or action — even if they don't say "Irasutoya" by name but clearly want a free, cute illustration for a slide, 발표자료, blog, doc, or chat. Also use it to optimize vague requests into fast Japanese keyword candidates and to return page/image URLs. Do NOT use it to generate brand-new AI art (e.g. DALL·E, Midjourney), draw charts or diagrams, design logos or icons, cut out/edit existing images, or recommend other stock-image sites — it only searches Irasutoya's existing catalog.
In this repository6 other files it gives its agents.
- AGENTS.mdAGENTS.md
- commands-creatorSkill
- herdrSkill
- install-skillSkill
- prototype-deploySkill
- visual-companionSkill
The file
Show its rules and text
We split this skill into the rules it gives an agent
19 rules. About 0 are common, 10 are distinctive, and 9 are very specific to this project.
≈0
Common · estimate
Many files like this have a rule of this kind.
≈10
Distinctive · estimate
This author's own choices. Worth a look.
9
Very specific
Tied to their project's names, files or tools. Probably skip.
- Common
- Distinctive
- Very specific
- Very specific: line 10
Use the installed irasutoya command-line tool to find Irasutoya illustrations.
- Very specific: line 14
Use irasutoya random when the user wants an arbitrary Irasutoya image.
- Very specific: line 15
Use irasutoya search "query" when the user gives a topic, keyword, or phrase.
- Very specific: line 16
Use irasutoya help or irasutoya help COMMAND when the user asks what is available.
- Distinctive: line 42
Search candidates in parallel when possible.
- Distinctive: line 64
Do not rely on long Korean natural-language strings as the only search.
- Very specific: line 79
Prefer 2-term Japanese queries such as 不思議 男の子, 疑問 男の子, 虫眼鏡 男の子.
- Distinctive: line 80
If a request includes a visible prop or action, search that prop/action plus subject before searching abstract emotions.
- Distinctive: line 81
If the first good result is a multi-image expression sheet, keep searching for a single-image scene when the user asked for a scene or full …
- Distinctive: line 91
Run exactly one refinement batch before reporting a partial match:
- Very specific: line 93
If there are no results, switch word form or synonym: 困る -> 困った, 会社員 -> サラリーマン.
- Distinctive: line 94
If results are expression sheets or grouped parts, add a concrete prop/action or try a more scene-like phrase.
- Very specific: line 95
If broad role + prop searches return only grouped images, demote them and search a visual proxy such as 白衣 マスク.
- Distinctive: line 96
If the exact desired result appears as the second result, select it over a broader first result and mention the selected title.
- Very specific: line 100
When comparing several candidate queries or checking whether the rubric holds, the inline image preview can flood the terminal.
- Distinctive: line 109
Use at least these smoke cases:
- Distinctive: line 138
Treat that warning as non-fatal.
- Very specific: line 144
If the CLI prints multiple Image URL lines for one page, identify the specific image that best matches instead of treating all images as …
- Distinctive: line 145
Do not scrape extra content unless the user asks for more than the CLI output.
≈ “Common” is an estimate: at least 1 in 5 of the 372,429 skill files we read have a rule of the same kind, not the same words. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.
Files
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 randomwhen the user wants an arbitrary Irasutoya image. - Use
irasutoya search "query"when the user gives a topic, keyword, or phrase. - Use
irasutoya helporirasutoya help COMMANDwhen the user asks what is available.
Expected commands:
irasutoya random
irasutoya search "cat"
irasutoya help searchThe search command returns up to 3 matching images. Each result prints:
Page URLTitleDescription- one or more
Image URLlines - an inline image preview when the terminal supports it
Fast Search Workflow
- 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
- Generate 2-4 concise query candidates. Prefer Japanese nouns and short noun phrases because Irasutoya titles/descriptions are Japanese.
- Search candidates in parallel when possible. Avoid launching many broad searches; stop once a strong match appears.
- Score results before reporting. Prefer results whose title or description matches at least two facets.
- 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.
Cap the score at 5. Use a result scoring 4-5 as the primary answer. Use a 3 only when no better result appears and explain the limitation briefly.
Query Strategy
- Do not rely on long Korean natural-language strings as the only search. Try the user's wording once only if it is likely indexed, then optimize.
- Convert Korean descriptions into Japanese visual keywords:
- curious/wondering:
不思議,疑問,はてな - boy:
男の子 - girl:
女の子 - child:
子供 - observe/look closely:
観察,見る - magnifying glass:
虫眼鏡 - question/question mark:
質問,疑問,はてな - angry:
怒る,怒った - crying:
泣く,泣いている - troubled/stuck:
困った,悩む - office worker:
会社員,サラリーマン - computer:
パソコン - doctor/lab coat:
医者,白衣
- curious/wondering:
- Prefer 2-term Japanese queries such as
不思議 男の子,疑問 男の子,虫眼鏡 男の子. - If a request includes a visible prop or action, search that prop/action plus subject before searching abstract emotions.
- If the first good result is a multi-image expression sheet, keep searching for a single-image scene when the user asked for a scene or full illustration.
- Try both dictionary-like and title-like forms for emotions/actions when the first form is weak:
怒る 女の子can beat怒った 女の子for a single angry girl.困った 会社員can beat困る 会社員for an office worker at a computer.泣いている 赤ちゃんcan surface the exact crying baby image, but check the second result when the first adds extra context.
- For occupational or role terms, try Irasutoya-style synonyms in the refinement batch, such as
会社員andサラリーマン. - Preserve the user's original intent in the final wording, not necessarily the original query text.
Failure Rules
Run exactly one refinement batch before reporting a partial match:
- If there are no results, switch word form or synonym:
困る->困った,会社員->サラリーマン. - If results are expression sheets or grouped parts, add a concrete prop/action or try a more scene-like phrase.
- If broad role + prop searches return only grouped images, demote them and search a visual proxy such as
白衣 マスク. - If the exact desired result appears as the second result, select it over a broader first result and mention the selected title.
Checking Candidate Queries (metadata only)
When comparing several candidate queries or checking whether the rubric holds, the inline image preview can flood the terminal. Suppress it by piping the CLI output through grep so only the metadata lines remain:
for q in "不思議 男の子" "虫眼鏡 男の子"; do
echo "=== $q ==="
irasutoya search "$q" | grep -E "Page URL|Title|Description|Image URL"
doneUse at least these smoke cases:
- curious boy wondering: expect
不思議 男の子or虫眼鏡 男の子to find a 5-point result. - angry girl: expect
怒る 女の子to beat怒った 女の子because the latter returns expression sheets. - office worker troubled at a computer: expect
困った 会社員to beat困る 会社員andパソコン 困った. - crying baby: expect
泣いている 赤ちゃんto include the exact simple crying baby image, even if it is not first.
The rubric is acceptable when at least three of four smoke cases find a 4-5 point result within the first refinement batch. If a case fails, add only the missing query synonym or failure rule; do not add broad generic search advice.
Running the CLI
The skill always uses the globally installed irasutoya executable:
irasutoya random
irasutoya search "query"Quote multi-word queries. If the irasutoya command is not found, install it with gem install irasutoya-cli (requires Ruby >= 3.1); the gem ships the irasutoya, terminal_image, and thor dependencies it needs.
Terminal Preview Behavior
Inline previews depend on terminal_image. If the terminal does not support inline images, the CLI still prints the page/image URLs and warns:
warn: This terminal is not able to show images inline
warn: Please use iTerm2 or terminal installed libsixel.Treat that warning as non-fatal. Return the printed URLs to the user when image rendering is unavailable.
Output Guidance
- Quote multi-word queries in shell commands.
- Report title, page URL, image URL, and a short reason why it matches.
- If the CLI prints multiple
Image URLlines for one page, identify the specific image that best matches instead of treating all images as equal. - Do not scrape extra content unless the user asks for more than the CLI output.
Does it work?
Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.
No reviews yet
People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.
- It worked
- Partly
- It didn't
Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.
Your trail
Files you open will collect here, so you can find them again.
Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.
Now do this for your own.
Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.

