Agent-Skills.md

Agent Skills: Eat Like Japanese Local Foodie

Research restaurants the way local Japanese foodies would. Use when Codex needs to find, compare, rank, or document what to eat in Japan for a trip, especially when the user asks for local Japanese recommendations, foodie picks, Tabelog, Hyakumeiten, Michelin Japan, Japanese beef, ramen, coffee, cafes, family-friendly meals, or Japan city/area food surveys.

UncategorizedID: chuyeow/agentic-layer/eat-like-japanese-local-foodie

Repository

chuyeow/agentic-layer
chuyeow
2

Install this agent skill to your local

pnpm dlx add-skill https://github.com/chuyeow/agentic-layer/tree/HEAD/skills/eat-like-japanese-local-foodie

Skill Files

Browse the full folder contents for eat-like-japanese-local-foodie.

Download Skill

Loading file tree…

skills/eat-like-japanese-local-foodie/SKILL.md

Skill Metadata

Name
eat-like-japanese-local-foodie
Description
Research restaurants the way local Japanese foodies would. Use when Codex needs to find, compare, rank, or document what to eat in Japan for a trip, especially when the user asks for local Japanese recommendations, foodie picks, Tabelog, Hyakumeiten, Michelin Japan, Japanese beef, ramen, coffee, cafes, family-friendly meals, or Japan city/area food surveys.

Eat Like Japanese Local Foodie

Research Japanese restaurants source-first, in Japanese where possible, then convert the result into practical trip decisions.

Core Workflow

  1. Define the eating context:

    • city, station/neighborhood, hotel anchor, travel dates
    • meal type: lunch, dinner, snack, cafe, takeaway
    • constraints: children, smoke-free, budget, reservation, walking distance, luggage, late arrival, dietary needs
    • intent: iconic local food, serious foodie meal, casual family meal, backup near route

  2. Search like a local:

    • Use Japanese station + genre searches before English searches.
    • Prefer station/neighborhood over city-level terms.
    • Search both quality and friction terms: 予約, 行列, 子連れ, 禁煙, 現金のみ, 売り切れ, 定休日.
    • Search negative terms when the candidate matters: まずい, 微妙, 観光客向け, 高すぎる.

  3. Build the candidate set:

    • Tabelog area rankings and genre rankings.
    • Tabelog Award / 百名店 / Hot Pot 100 / Ramen 100 / genre-specific 100 lists.
    • Michelin for fine dining only; do not treat it as the default local-foodie source.
    • Google Maps for recent logistics, not primary taste ranking.
    • Instagram/X for recency, queues, new openings, specials, and local buzz.
    • Japanese food blogs/local media for context and hidden practical details.
    • Reservation platforms: Tabelog, TableCheck, OMAKASE, Pocket Concierge, restaurant official site.

  4. Cross-check each serious candidate:

    • Tabelog score, review count, genre, awards, area ranking.
    • Google Maps rating, review count, latest reviews, language mix.
    • Recent hours/closed days from official site or recent Google/Tabelog data.
    • Menu/pricing: lunch vs dinner, courses, child pricing, cover charge, cash/card.
    • Reservation friction: walk-in, phone-only, online booking, cancellation policy.
    • Physical fit: counter/table/private room, stroller/luggage, smoke-free, child-friendly.
    • Route fit from trip anchor and likely day itinerary.
    • Photos: actual dish quality, menu board, queue, clientele, seating, portion size.

  5. Rank for the user’s trip, not abstract quality:

    • Go: high confidence and fits itinerary.
    • Maybe: good but has friction, cost, distance, or duplicate role.
    • Backup: useful near route, lower stakes, logistics-first.
    • Skip: weak signal, tourist-trap signal, too far, closed, not child-fit, or redundant.

Japanese Search Patterns

Use combinations like:

<station> <genre> 名店
<station> <genre> 食べログ
<station> <genre> 百名店
<station> <genre> 子連れ
<station> <genre> 予約
<station> <genre> 行列
<station> <genre> ランチ
<station> <genre> ディナー
<station> <genre> 穴場
<station> <genre> 老舗
<station> <genre> 地元
<restaurant name> 評判
<restaurant name> 口コミ
<restaurant name> 予約困難
<restaurant name> 子連れ
<restaurant name> まずい
<restaurant name> 現金のみ

Useful Osaka area terms:

なんば
難波
日本橋
裏なんば
法善寺横丁
千日前
道頓堀
心斎橋
梅田
天満
福島

Useful genre terms:

お好み焼き
たこ焼き
うどん
焼肉
ホルモン
すき焼き
しゃぶしゃぶ
串カツ
ラーメン
カレー
寿司
天ぷら
居酒屋
喫茶店
カフェ
和菓子

Useful social searches:

#大阪グルメ
#難波グルメ
#裏なんば
#関西グルメ
#大阪ランチ
#大阪ディナー
#大阪カフェ
#食べログ百名店

How Foodies Actually Behave

  • They start from a dish or neighborhood, not “best restaurants in Osaka”.
  • They keep separate lists for serious meals, casual meals, snacks, backups, and late-night options.
  • They care about timing: when queues form, when popular items sell out, lunch value, last order, and closed days.
  • They read bad reviews deliberately to understand failure modes.
  • They look at reviewer credibility and photo evidence, not just star ratings.
  • They use Tabelog for Japanese taste signal and Google Maps for current logistics.
  • They know a 3.4 casual shop on Tabelog can be excellent; 3.5+ is strong; 4.0+ is elite.
  • They do not assume Michelin equals best local meal; it is more useful for expensive destination dining.
  • They check whether a place is popular with locals, tourists, office workers, students, families, or influencers.
  • They check the ordering system before going: ticket machine, QR order, course only, one-drink rule, cash-only.
  • They choose by route fit. A good restaurant near the plan usually beats a great restaurant that breaks the day.
  • They avoid overfitting one ranking. Consensus across Tabelog, recent Japanese reviews, photos, and route fit matters.

Tourist-Trap Filters

Flag and de-prioritize when several are true:

  • Very high Google score but low/flat Tabelog signal.
  • Review base dominated by first-time foreign tourists.
  • Generic “Kobe beef / wagyu experience” language with weak Japanese local reviews.
  • Aggressive multilingual street signage, touting, laminated mega-menu, or all-in-one “Japan food” offer.
  • Recent reviews mention rushed service, bait pricing, forced courses, hidden fees, or poor value.
  • Photos show style/novelty but not food quality.
  • Location is prime tourist corridor with no local-review support.

Do not automatically skip tourist-friendly places. They can be correct when the user needs English, easy booking, predictable child fit, or low friction. Label them honestly.

Output Shape

For research notes, prefer:

  1. Short answer / ranked shortlist.
  2. Decision table:
    • priority
    • place
    • area
    • genre
    • signal
    • route/logistics
    • practical read
  3. Detail cards only for serious candidates:
    • why go
    • what to order
    • avoid/risks
    • reservation/queue
    • child/logistics
    • map link
    • source links
  4. “How to use this list” section:
    • where to book
    • which day/route it fits
    • what to do if queue is bad
    • which candidates are backups

When saving into a trip repo, put the note near the relevant trip folder and link it from the trip index. Keep source links in the note.

Source Standards

  • Browse for current restaurant data. Hours, rankings, closures, prices, and reviews change.
  • Prefer official restaurant/Tabelog/TableCheck pages for hours, pricing, reservations, and closures.
  • Use Reddit/social/blogs as color, not sole evidence.
  • Never invent exact prices, opening hours, awards, or closed days.
  • If a signal is conflicting, state the conflict and mark it for verification.
  • Do not quote long reviews. Summarize patterns and link sources.