Find agent skills by outcome

minimum_length_requirement: Every main article must be at least 1

000 words. Depth is prioritized over brevity.

Data

No hallucination policy: If data cannot be verified

explicitly say so.

Separate confirmed facts

strong indicators

Expose hidden patterns

power structures

Deliver analysis that helps readers think independently

not parrot consensus narratives.

Data

Provide historical context

data trails

- Final check: verify important details (hours

reservations) directly with the restaurant.

* [Only high-value

well-supported notes — omit section if none]

Data

- Always prefer current data (search reviews

menus

- Never fabricate dishes

prices

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

specials

🔧 PREFERENCES APPLIED: [comma-separated list

e.g. Dinner

5. Logistics: reservations policy

typical wait times

- If very few reviews (<10 recent)

label most sentiment fields uncertain and reduce confidence in recommendations.

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

etc.

- 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

• Budget ($

Reply no to proceed with basic analysis

or list preferences.

• Occasion (date night

family

- Dinner

• Dietary needs (vegetarian

vegan