ScreenSense Research Swarm

Pattern: Parallel Workers | Team size: 5 agents

This team investigates the same media title from multiple angles at the same time, allowing quick and well-rounded label generation. Parallel workers fit because popular children's media can be evaluated independently across pedagogy, engagement design, business model, and parent sentiment before being merged into one score.

Business Challenge

The Screen-Time "Nutritional Label" Generator — A tool that generates a "Nutrition Label" for popular children's media and games. Parents type in an app or show name, and the tool generates a breakdown of "Ingredients": % Passive Watching, % Active Logic, Sugar/Dopamine level (high/low), Artificial Additives (Ads/In-app purchases). — Highly shareable visual format that instantly communicates the difference between "junk" screen time and "enriching" screen time.

Agent Roles

Generated Prompt

Create an agent team to generate a parent-facing Screen-Time “Nutritional Label” for a children’s app, game, YouTube channel, streaming show, or digital media title. Use the ScreenSense Research Swarm pattern: five parallel specialist agents independently evaluate the same title from different angles, then merge their findings into one clear, shareable label that helps parents distinguish “junk” screen time from enriching screen time.

Project name: screensense_research_swarm

Primary output directory: outputs/agent_teams_demo/screensense_research_swarm/

Core mission:
Build a concise but well-researched “nutrition label” for one children’s media title. The label must translate research into parent-friendly categories:
- Passive Watching %
- Active Logic %
- Creative/Curiosity %
- Social-Emotional Value %
- Sugar/Dopamine Level: Low / Medium / High / Extreme
- Artificial Additives: Ads, in-app purchases, subscriptions, branded content, data collection, algorithmic recommendations, autoplay, loot boxes, etc.
- Age Fit
- Parent Takeaway
- Better Use Tips

If the user has not provided a title, immediately ask for:
1. Media title
2. Media type: app, game, streaming show, YouTube channel, website, console game, other
3. Target child age or grade
4. Country/region, if monetization or app store information may vary
5. Any parent concern to prioritize, such as ads, addictive design, educational value, violence, social features, or cost

Do not begin the analysis until the title is known.

Research standards:
- Use current public information where available: official product pages, app store listings, Common Sense Media, parent reviews, educator reviews, privacy labels, news coverage, developer/publisher documentation, platform pages, and reputable review sites.
- Cite URLs or source names in each agent file when available.
- If web access is unavailable, clearly mark assumptions and use only general knowledge.
- Do not fabricate precise statistics. Use estimated percentages only after evidence review and label them as “ScreenSense estimate.”
- Do not make medical or diagnostic claims. “Sugar/Dopamine Level” is a parent-facing metaphor for reward intensity, not a clinical measure.
- Distinguish between content quality, engagement design, and monetization model.
- Treat different versions separately when relevant, such as free vs paid app, YouTube vs Netflix version, mobile vs console game.
- Use direct, plain English suitable for parents of children ages 3–12.

Required workflow and dependencies:

Step 0 — Intake and shared rubric setup
This task must complete before any specialist agent begins.
Create outputs/agent_teams_demo/screensense_research_swarm/00_intake_and_rubric.md with exactly these sections:

# Intake
- Title:
- Media type:
- Publisher/creator/platform:
- Target age requested:
- Region:
- Version or access model:
- Date of review:

# Known Constraints
List missing information, unclear versions, or research limitations in 5–10 bullets.

# Shared Scoring Rubric
Define the common scoring scale all agents must use:
- 0 = absent or harmful
- 1 = weak
- 2 = mixed or limited
- 3 = solid
- 4 = strong
- 5 = exceptional

Define ScreenSense ingredient categories:
- Passive Watching: non-interactive consumption, repetitive viewing, background watching, autoplay-driven viewing
- Active Logic: problem-solving, strategy, puzzles, cause/effect reasoning, math, planning
- Creative/Curiosity: making, experimenting, storytelling, exploration, open-ended play
- Social-Emotional: empathy, emotional naming, cooperation, self-regulation, prosocial modeling
- Sugar/Dopamine: reward frequency, variable rewards, rapid pacing, streaks, cliffhangers, surprise reveals, autoplay, notifications
- Artificial Additives: ads, sponsorships, in-app purchases, subscriptions, branded content, loot boxes, data-driven targeting, algorithmic funnels

# Collaboration Rules
Include these rules:
- Each specialist must write an independent first-pass assessment before reading other specialists’ conclusions.
- Each specialist must include “Assumptions to Challenge” at the end of their file.
- After all five specialist files exist, agents must read each other’s summaries and add disagreements or confirmations to the cross-agent challenge log.
- The synthesizer must not average scores mechanically; it must resolve conflicts using evidence quality and parent relevance.
- If two agents disagree, document the disagreement and make a final judgment.

Step 1 — Parallel specialist analysis
After Step 0 is complete, run these five agents in parallel. Each agent must investigate the same title independently and write its own file before synthesis begins.

Agent 1: Educational Value Assessor
Role:
Evaluate whether the media promotes literacy, numeracy, problem-solving, emotional learning, creativity, curiosity, persistence, or knowledge acquisition.

Output file:
outputs/agent_teams_demo/screensense_research_swarm/01_educational_value_assessor.md

Required structure and length: 700–1,000 words.

Use exactly these headings:
# Educational Value Assessment
## 1. Executive Finding
Write 80–120 words with the main educational judgment.

## 2. Learning Domains Detected
Use a table with columns:
- Domain
- Evidence
- Strength Score 0–5
- Parent Interpretation

Include rows for:
- Literacy/Language
- Numeracy/Math
- Problem-Solving/Logic
- Creativity/Open-Ended Play
- Curiosity/Knowledge Building
- Social-Emotional Learning
- Motor/Spatial Skills, if relevant

## 3. Active vs Passive Learning
Estimate the share of use that is active learning vs passive exposure. Provide 3–6 bullets explaining the estimate.

## 4. Depth vs Surface Learning
Explain whether the media encourages deep reasoning, repeat practice, memorization, imitation, or shallow engagement.

## 5. Best-Case Use
Describe how a parent or child could use the media in the most enriching way.

## 6. Educational Risks or Overclaims
Call out weak curriculum claims, “edutainment” overstatements, repetitive drills, or misleading learning claims.

## 7. Evidence and Sources
List source names and URLs where available. If sources are unavailable, say so.

## 8. Assumptions to Challenge
List 3–5 assumptions another agent should challenge.

Agent 2: Engagement Mechanics Analyst
Role:
Assess pacing, reward frequency, autoplay, streaks, variable rewards, surprise rewards, notifications, cliffhangers, binge design, short-loop gameplay, and other dopamine-triggering mechanics.

Output file:
outputs/agent_teams_demo/screensense_research_swarm/02_engagement_mechanics_analyst.md

Required structure and length: 700–1,000 words.

Use exactly these headings:
# Engagement Mechanics Assessment
## 1. Executive Finding
Write 80–120 words summarizing how “sticky” or habit-forming the media appears.

## 2. Reward and Pacing Inventory
Use a table with columns:
- Mechanic
- Present? Yes/No/Unknown
- Evidence
- Intensity: Low/Medium/High
- Parent Meaning

Evaluate at minimum:
- Autoplay or next-episode funnel
- Streaks/daily rewards
- Surprise or variable rewards
- Rapid cuts or high sensory pacing
- Leveling/badges/points
- Notifications or reminders
- Cliffhangers
- Infinite feed or algorithmic recommendations
- Short retry loops
- Social comparison or leaderboards

## 3. Sugar/Dopamine Level Recommendation
Assign Low, Medium, High, or Extreme. Explain in 150–220 words.

## 4. Friction to Stop
Assess how easy it is for a child to stop after 10, 20, or 30 minutes.

## 5. Parent Control Levers
List available settings or practical controls: timers, disabling autoplay, offline mode, ad-free subscription, episode selection, guided play.

## 6. Red Flags
List 3–8 engagement red flags.

## 7. Evidence and Sources
List sources and URLs where available.

## 8. Assumptions to Challenge
List 3–5 assumptions another agent should challenge.

Agent 3: Advertising and Purchase Scanner
Role:
Identify advertisements, branded content, subscriptions, in-app purchases, upsells, data-driven monetization risks, product placement, loot boxes, premium currency, and privacy concerns.

Output file:
outputs/agent_teams_demo/screensense_research_swarm/03_advertising_purchase_scanner.md

Required structure and length: 700–1,000 words.

Use exactly these headings:
# Advertising and Purchase Assessment
## 1. Executive Finding
Write 80–120 words summarizing monetization and commercial exposure risk.

## 2. Artificial Additives Inventory
Use a table with columns:
- Additive
- Present? Yes/No/Unknown
- Evidence
- Risk Level: Low/Medium/High
- Parent Meaning

Evaluate at minimum:
- Display/video ads
- Sponsored or branded content
- In-app purchases
- Subscriptions
- Upsells to premium
- Loot boxes/gacha/random purchases
- Premium currency
- Merchandise tie-ins
- Data collection or tracking
- Cross-promotion to other apps/videos
- External links or web browsing

## 3. Free vs Paid Experience
Compare free, paid, subscription, ad-supported, and premium versions if relevant.

## 4. Child Pressure Points
Explain where the design may create pester power, purchase pressure, or brand attachment.

## 5. Privacy and Data Notes
Summarize known privacy labels, child data handling, account requirements, or unknowns.

## 6. Artificial Additives Rating
Assign Clean, Light, Moderate, Heavy, or Unknown. Explain in 120–180 words.

## 7. Evidence and Sources
List app store pages, privacy pages, review sources, official pages, or unavailable source notes.

## 8. Assumptions to Challenge
List 3–5 assumptions another agent should challenge.

Agent 4: Age-Fit and Safety Reviewer
Role:
Judge whether themes, interaction style, reading level, complexity, social features, violence, fear, language, sexual content, stereotypes, chat, user-generated content, and safety controls are appropriate for the target age range.

Output file:
outputs/agent_teams_demo/screensense_research_swarm/04_age_fit_safety_reviewer.md

Required structure and length: 700–1,000 words.

Use exactly these headings:
# Age-Fit and Safety Assessment
## 1. Executive Finding
Write 80–120 words stating whether the title fits the requested age and under what conditions.

## 2. Age-Fit Table
Use a table with columns:
- Factor
- Evidence
- Fit for Requested Age: Poor/Mixed/Good/Excellent/Unknown
- Parent Meaning

Evaluate at minimum:
- Reading level
- Instructions and navigation
- Emotional intensity
- Violence/fear/scary imagery
- Language or crude humor
- Sexual/romantic content, if any
- Social interaction or chat
- User-generated content
- Online multiplayer
- Fine motor/cognitive demands
- Parent controls

## 3. Recommended Age Band
Give a recommended age band and explain any mismatch with official ratings.

## 4. Safety Conditions
List conditions that improve safety, such as co-viewing, disabling chat, using a child profile, turning off autoplay, avoiding public servers, or selecting specific episodes.

## 5. Watch-Outs by Age
Give practical notes for ages 3–5, 6–8, 9–12, and 13+.

## 6. Evidence and Sources
List sources and URLs where available.

## 7. Assumptions to Challenge
List 3–5 assumptions another agent should challenge.

Agent 5: Parent Sentiment Summarizer
Role:
Synthesize common parent reviews, educator opinions, public complaints, praise patterns, and practical context. Focus on what parents repeatedly notice after real household use.

Output file:
outputs/agent_teams_demo/screensense_research_swarm/05_parent_sentiment_summarizer.md

Required structure and length: 700–1,000 words.

Use exactly these headings:
# Parent Sentiment Assessment
## 1. Executive Finding
Write 80–120 words summarizing the overall parent/educator sentiment.

## 2. Common Praise Themes
List 5–8 bullets. Each bullet must include what parents like and why it matters.

## 3. Common Complaint Themes
List 5–8 bullets. Each bullet must include what parents dislike and the likely household impact.

## 4. Review Pattern Table
Use a table with columns:
- Theme
- Sentiment: Positive/Mixed/Negative
- Frequency Impression: Rare/Occasional/Common/Very Common/Unknown
- Example Concern or Praise
- Parent Meaning

## 5. Educator or Expert View
Summarize educator, child-development, or reputable review-site perspectives if available.

## 6. Household Fit
Describe which families are most likely to value this title and which may dislike it.

## 7. Evidence and Sources
List source names, review platforms, URLs where available, and note if review data is limited.

## 8. Assumptions to Challenge
List 3–5 assumptions another agent should challenge.

Step 2 — Cross-agent challenge and calibration
This task must begin only after all five specialist files are complete.
Create outputs/agent_teams_demo/screensense_research_swarm/06_cross_agent_challenge_log.md.

Required structure and length: 500–800 words.

Use exactly these headings:
# Cross-Agent Challenge Log
## 1. Key Agreements
List 5–10 bullets where multiple agents independently reached the same conclusion.

## 2. Key Disagreements or Tensions
Use a table with columns:
- Issue
- Agents Involved
- Disagreement
- Evidence on Each Side
- Resolution Needed for Synthesis

Include tensions such as:
- Educational value vs addictive mechanics
- Parent praise vs commercial risk
- Official age rating vs practical developmental fit
- Passive entertainment vs active interaction
- Free access vs monetization exposure

## 3. Assumptions Challenged
List at least 8 assumptions from specialist reports and whether they were upheld, revised, or rejected.

## 4. Calibration Notes for Synthesizer
Give 5–8 direct instructions to the synthesizer on how to resolve the final label.

Collaboration mechanics:
- Each agent must read the other four executive findings after finishing its own report.
- Each agent must contribute at least one challenge to the log.
- Challenges must be evidence-based, not stylistic.
- If evidence is weak or missing, mark the issue “Unknown” instead of forcing certainty.
- The synthesizer must use this challenge log before writing the final label.

Step 3 — Final Screen-Time Nutritional Label synthesis
This task must begin only after outputs/agent_teams_demo/screensense_research_swarm/06_cross_agent_challenge_log.md is complete.
Create outputs/agent_teams_demo/screensense_research_swarm/07_screen_time_nutritional_label.md.

Required structure and length: 900–1,300 words.

Use exactly these headings:

# Screen-Time Nutritional Label: [Title]

## 1. Front-of-Pack Verdict
Provide:
- Overall Rating: Enriching / Mostly Enriching / Mixed Snack / Junky Treat / Avoid or Heavily Supervise
- Best For: age band
- Use Style: solo / co-play / co-view / parent-guided / limited treat
- Session Size: recommended minutes per session
- Parent One-Liner: one sentence under 25 words

## 2. Nutrition Facts Panel
Use this exact markdown table:

| Ingredient | ScreenSense Estimate | What It Means |
|---|---:|---|
| Passive Watching | X% |  |
| Active Logic | X% |  |
| Creative/Curiosity | X% |  |
| Social-Emotional Value | X% |  |
| Commercial Additives | Low/Medium/High/Unknown |  |
| Sugar/Dopamine Level | Low/Medium/High/Extreme |  |
| Age Fit | Poor/Mixed/Good/Excellent |  |

Rules:
- The four percentage rows must total 100%.
- These are interpretive estimates, not measured scientific quantities.
- Explain any uncertainty in the “What It Means” column.

## 3. Ingredients List
Write a parent-friendly ingredients list in 8–12 bullets. Use food-label style language, such as:
- “Main ingredient: rapid-fire rewards and short challenge loops”
- “Contains: genuine puzzle-solving”
- “May contain: ads, upsells, or branded characters”
- “Fortified with: vocabulary exposure”

## 4. Sugar/Dopamine Explanation
Explain the Sugar/Dopamine level in 120–180 words. Mention the specific mechanics that drove the rating.

## 5. Artificial Additives Explanation
Explain monetization and commercial exposure in 120–180 words.

## 6. Educational Value Explanation
Explain the strongest and weakest learning elements in 150–220 words.

## 7. Age-Fit and Safety Notes
Give practical age guidance in 120–180 words.

## 8. Parent Sentiment Snapshot
Summarize parent/educator sentiment in 100–150 words.

## 9. Best-Case Use Recipe
Give 5–7 bullets that help parents turn this into healthier screen time. Include limits, co-use prompts, settings, and transition strategies.

## 10. Red Flag Checklist
Use checkboxes with 8–12 items. Example:
- [ ] Autoplay or infinite feed
- [ ] Ads or sponsorships
- [ ] In-app purchases
- [ ] Hard-to-stop reward loops

## 11. Better Alternatives or Pairings
Suggest 3–5 types of alternatives or pairings, not necessarily specific brands, such as “offline building activity after play,” “co-viewed documentary,” or “open-ended drawing app.” Explain each in one sentence.

## 12. Evidence Base
List the specialist files used and the top 5–10 external sources or source categories. Include uncertainty notes.

Synthesis instructions:
- Integrate all five specialist reports and the cross-agent challenge log.
- Do not simply average scores. Weight evidence in this order: direct product evidence, official monetization/privacy information, reputable third-party reviews, parent review patterns, general category knowledge.
- If educational value is high but engagement mechanics are intense, classify the final rating as mixed rather than purely enriching.
- If monetization risk is heavy for the target age, reflect that prominently even if content quality is good.
- If parent sentiment conflicts with expert review, explain the reason rather than hiding the conflict.
- Keep the tone useful, not alarmist.
- Make the final label highly shareable: short phrases, memorable ingredient wording, and clear parent action steps.

Step 4 — Final review and quality gate
This task must begin only after outputs/agent_teams_demo/screensense_research_swarm/07_screen_time_nutritional_label.md is complete.
Create outputs/agent_teams_demo/screensense_research_swarm/08_final_review.md.

Required structure and length: 500–800 words.

Use exactly these headings:
# Final Review and Quality Gate
## 1. Completeness Check
Confirm whether all required files exist:
- 00_intake_and_rubric.md
- 01_educational_value_assessor.md
- 02_engagement_mechanics_analyst.md
- 03_advertising_purchase_scanner.md
- 04_age_fit_safety_reviewer.md
- 05_parent_sentiment_summarizer.md
- 06_cross_agent_challenge_log.md
- 07_screen_time_nutritional_label.md

## 2. Rubric Consistency Check
Verify that ratings and terminology are consistent across files. Note any mismatches and corrections.

## 3. Evidence Quality Check
Rate evidence quality as Strong, Moderate, Limited, or Weak. Explain why in 100–150 words.

## 4. Parent Usefulness Check
Evaluate whether the final label answers:
- Is this enriching or junky?
- How addictive is it?
- Are there ads or purchases?
- Is it age-appropriate?
- How should parents use it?

## 5. Risk and Fairness Check
Confirm:
- No fabricated claims
- No clinical claims about dopamine or addiction
- Unknowns are clearly marked
- Positive and negative evidence are both represented
- The review distinguishes content quality from business model
- The recommended age guidance is practical

## 6. Required Revisions
List any final revisions needed. If none, write “No required revisions.”

## 7. Final Approval
End with exactly one of:
- APPROVED FOR PARENT-FACING USE
- APPROVED WITH UNCERTAINTY NOTES
- NOT APPROVED — REVISIONS REQUIRED

Global formatting requirements:
- All files must be markdown.
- Use concise parent-friendly language.
- Avoid academic jargon unless immediately explained.
- Use tables where required.
- Keep all claims traceable to either evidence, specialist judgment, or explicit uncertainty.
- Use “ScreenSense estimate” for any percentage or interpretive rating.
- Do not include hidden reasoning or private chain-of-thought. Provide clear conclusions and evidence summaries only.

Final synthesis/review requirement:
The run is not complete until 08_final_review.md exists and gives an approval status. If the quality gate finds missing evidence, inconsistent ratings, or unsupported claims, revise 07_screen_time_nutritional_label.md before final approval and document the revision in 08_final_review.md.

Build Your Own

Create your own AI agent team at Build Agents Store. Describe your business problem and get specialized agent teams with ready-to-use prompts for Claude Code.