Find agent skills by outcome

No hallucination policy: If data cannot be verified

explicitly say so.

Writing & Content

minimum_length_requirement: Every main article must be at least 1

000 words. Depth is prioritized over brevity.

Data

Provide historical context

data trails

Deliver analysis that helps readers think independently

not parrot consensus narratives.

Expose hidden patterns

power structures

- Final check: verify important details (hours

reservations) directly with the restaurant.

- Never fabricate dishes

prices

Data

- Always prefer current data (search reviews

menus

* [Only high-value

well-supported notes — omit section if none]

🔧 PREFERENCES APPLIED: [comma-separated list

e.g. Dinner

7. Extras: only include well-supported notes (happy hour

specials

5. Logistics: reservations policy

typical wait times

- For occasion: date night → intimate/shareable/romantic plates; family → generous portions/kid-friendly; celebration → impressive/specials

etc.

- If very few reviews (<10 recent)

label most sentiment fields uncertain and reduce confidence in recommendations.

- Tailor to dietary_preferences

preferred_meal_type

One-shot / non-interactive fallback: If this is a single message or preferences are not provided

assume no and proceed directly to core analysis.

2. Collect & summarize recent reviews (Google

Yelp

- If multiple similar restaurants exist

state which one is selected and why (e.g. highest review count

Reply no to proceed with basic analysis

or list preferences.

• Occasion (date night

family

• Budget ($

- Vegan

brunch

- Dinner

• Dietary needs (vegetarian

vegan