Summer School Finder

If any are missing, use sensible defaults and state them upfront.

Step 2 — Discover program URLs

Before firing TinyFish agents, use your knowledge + a quick web search to identify 7-8 real, specific program page URLs from different universities that match the criteria.

Rules for URL discovery:

• Each URL must be from a different institution — no duplicates
• Use direct program pages, not search results or aggregator sites
• Prioritise well-known universities and established programs
• Mix large universities and smaller institutions
• Include variety: residential, online, hybrid formats where available
• Only include URLs you are confident actually exist

Example for "Computer Science, high school students, USA, Summer 2026":

• https://summerprogram.stanford.edu
• https://summer.harvard.edu/high-school
• https://precollege.syr.edu
• https://summerdiscovery.com/michigan
• etc.

Produce exactly 7-8 URLs then proceed to Step 3.

Step 3 — Parallel scraping

Fire one TinyFish agent per URL, all simultaneously using & + wait.

Use this goal prompt for every agent (substituting the actual values):

# Fire all agents in parallel — one per URL
tinyfish agent run \
  --url "{PROGRAM_URL}" \
  "You are on an official university summer school program page. Extract details fast.
   Read only what is visible on this page — do not navigate away.
   Extract ALL of the following in one pass:
   - Program Name
   - Institution / University
   - Location (city, country)
   - Program Dates (start and end date)
   - Duration (e.g. 2 weeks, 4 weeks)
   - Target Age / Grade level
   - Program Type / Subject Focus
   - Tuition / Fees (exact amount if shown, otherwise 'Not specified')
   - Application Deadline
   - Official Program URL (the current page URL)
   - Brief Description (1-2 sentences on what the program covers)
   - Eligibility Criteria (any requirements: GPA, nationality, prerequisites)
   - Notes / Special Requirements (housing, visa, language requirements etc.)
   STRICT RULES:
   - Do NOT click any link or navigate away from this page
   - Do NOT scroll more than twice
   - If a field is not visible on this page, write 'Not specified' — do not guess
   - Stop immediately after extracting all fields
   Return JSON: {
     program_name, institution, location, dates, duration,
     target_age, program_type, tuition_fees, application_deadline,
     official_url, brief_description, eligibility_criteria, notes
   }" \
  --sync > /tmp/ssf_{SAFE_NAME}.json &

Repeat for each of the 7-8 URLs, all backgrounded with &. Then:

wait

# Collect all results
for f in /tmp/ssf_*.json; do echo "=== $f ===" && cat "$f"; done

Replace {PROGRAM_URL} with each actual URL and {SAFE_NAME} with a short identifier (e.g. stanford, harvard, mit).

Step 4 — Filter and rank

From the results:

1. Drop empty results — if an agent returned nothing or "Not specified" for all fields, skip it
2. Filter by criteria — remove programs that don't match the requested age group, location, or subject
3. Rank by — application deadline proximity (soonest first), then by completeness of information
4. Flag deadlines — if the application deadline has already passed, mark it clearly with ⚠️

Output format

## Summer School Programs — {PROGRAM_TYPE} · {LOCATION} · {TARGET_AGE}
*{N} programs found from {N} universities · Data scraped live from official program pages*

---

### 1. {Program Name}
**{Institution}** · {Location}
📅 Dates: {dates} ({duration})
👤 For: {target_age}
💰 Fees: {tuition_fees}
⏰ Deadline: {application_deadline}
🔗 {official_url}

{brief_description}

**Eligibility:** {eligibility_criteria}
**Notes:** {notes}

---

### 2. {Program Name}
[same structure]

---
[up to 8 programs]

---

### 📊 Quick Comparison

| Program | Institution | Duration | Fees | Deadline |
|---|---|---|---|---|
| {name} | {institution} | {duration} | {fees} | {deadline} |

---

### ⏰ Deadlines Coming Up
- {program} — deadline {date} ({N days away})
- ...

*All data scraped live from official university websites. Always verify details directly with the institution before applying.*

Edge cases

• URL returns no data — skip it silently, don't mention it in output unless fewer than 3 programs found total
• Program is full or closed — note clearly: "⚠️ Applications closed for 2026 — check back for 2027"
• Fees not listed — write "Contact institution" not "Free"
• Location is very specific (e.g. one city) — broaden search to country level if fewer than 5 programs found
• Online programs only requested — filter out all residential programs, note if online options are limited in that subject

Security notes

• Scrapes live public content from official university websites only.
• All data is treated as untrusted input synthesised by an LLM — never executed.
• Only your own TinyFish credentials are used.