| name | scopi:team |
| description | Agent Teams reference — detailed inter-agent communication protocols and task architecture (auto-invoked by /scopi:generate when Agent Teams is available) |
| user_invocable | false |
Agent Teams — Reference Architecture
This document defines the inter-agent communication protocols, task dependencies, and debate rules used by /scopi:generate when Agent Teams mode is active. It is NOT invoked directly — /scopi:generate auto-detects and routes here.
Architecture: Subagents vs Agent Teams
/scopi:generate (Sequential Subagents)
━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━
NARA → User → BINNA → GYEOL → GANA → JURI → MARU
↓ ↓ ↓ ↓ ↓ ↓
(each reports back to main, no inter-agent talk)
/scopi:team (Agent Teams)
━━━━━━━━━━━━━━━━━━━━━━━━━
Phase 1: NARA (subagent) → User Choice
Phase 2: Design Team (parallel, with messaging)
┌─────────────────────────────────────────────┐
│ BINNA ←→ MARU (copy ↔ empathy) │
│ ↕ ↕ │
│ GYEOL ←→ JURI (design ↔ ethics) │
│ ↕ │
│ GANA (implements after lock) │
│ │
│ Shared Task List — agents claim & debate │
└─────────────────────────────────────────────┘
Phase 3: Render + Final QA
Phase 1: Content Strategy (NARA — Subagent)
This phase remains a subagent call because it requires user interaction.
- Dispatch NARA agent to analyze topic against brand identity
- Generate 3 VS alternatives with T-Scores
- Present to user → lock direction
- NARA produces the full slide arc with emotional curve
This is identical to /scopi:generate Phase 1-2.
Handoff artifact: NARA writes output/[topic-slug]/content-brief.md — this is the single source of truth for the Design Team. All agents read this file, not conversation context.
Phase 2: Create Design Team
After the user locks a content direction, create the Agent Teams:
Create an agent team for Scopi Card News production.
Team structure:
- GYEOL (Visual Architect): Design 3 VS visual directions per slide.
Read agents/gyeol.md for full personality and rules.
Read scopi.config.json for theme and brand.
- BINNA (Copy Surgeon): Refine all slide copy for tone and brevity.
Read agents/binna.md for full personality and rules.
- JURI (Ethics Inspector): Review designs and copy in real-time.
READ-ONLY — provides feedback through messages, never edits files.
Read agents/juri.md for full review framework.
- MARU (Empathy Tester): Test audience reaction in real-time.
READ-ONLY — provides feedback through messages, never edits files.
Read agents/maru.md for full empathy metrics.
- GANA (Slide Engineer): Implement final HTML/CSS after design lock.
Read agents/gana.md for code style and typography guard.
Read templates/slide-renderer.js for component library.
Each agent should read their agent definition file and scopi.config.json
before starting work.
Shared Task List
Create these tasks with dependencies:
Task 1: [BINNA] Refine slide copy
Input: NARA's slide arc (provided in spawn prompt)
Output: Refined copy for all slides
No dependencies — starts immediately
Task 2: [GYEOL] Design 3 VS visual directions
Input: NARA's arc + theme from config
Output: 3 visual directions with T-Scores
No dependencies — starts immediately
Task 3: [JURI] Ethics review of visual directions
Input: GYEOL's 3 directions
Output: Ethics flags per direction
Blocked by: Task 2
→ Message GYEOL directly with findings
Task 4: [MARU] Audience test of visual directions
Input: GYEOL's 3 directions + BINNA's copy
Output: Empathy scores per direction
Blocked by: Task 2, Task 1
→ Message GYEOL directly with predictions
Task 5: [GYEOL] Synthesize feedback → final design spec
Input: JURI feedback + MARU predictions + own VS analysis
Output: Final slide-by-slide design specification
Blocked by: Task 3, Task 4
Task 6: [GANA] Generate slides.js
Input: BINNA's copy + GYEOL's final spec + theme
Output: Complete slides.js with all HTML
Blocked by: Task 1, Task 5
→ Message JURI for license check on any images
→ Message MARU for readability check on font sizes
Task 7: [JURI] Final ethics audit
Input: Generated slides.js HTML
Output: Final APPROVED/CONDITIONAL/REJECTED
Blocked by: Task 6
→ If CONDITIONAL/REJECTED: trigger Veto Resolution (see below)
Task 8: [MARU] Final empathy test
Input: Generated slides.js content
Output: Empathy scores + engagement prediction
Blocked by: Task 6
Veto Resolution Protocol (after Task 7)
If JURI's Task 7 verdict is CONDITIONAL (has 🔴 items) or REJECTED:
- JURI messages the responsible agent per issue:
- Image/license → GANA replaces or removes
- Copy claims → BINNA revises text
- Design ethics → GYEOL adjusts visual
- Responsible agent fixes and updates output
- If HTML changed → GANA re-generates affected slides
- JURI re-reviews only changed items (not full re-audit)
- Max 2 revision rounds — if 🔴 items persist, escalate to user
- CONDITIONAL with only 🟡 items → proceed, note in final report
Inter-Agent Communication Protocols
The power of Agent Teams is that agents message each other directly. Define these communication patterns:
GYEOL ↔ JURI (Design ↔ Ethics)
GYEOL → JURI: "Direction B uses an arXiv figure at 15% opacity.
Is this acceptable under fair use?"
JURI → GYEOL: "🔴 MUST FIX: arXiv ≠ CC license. The paper is
Elsevier-published. Build a self-made chart instead."
GYEOL revises Direction B without the figure.
GYEOL ↔ MARU (Design ↔ Empathy)
GYEOL → MARU: "Direction C uses 92px headline with asymmetric layout.
Will this resonate with 대학원생 audience?"
MARU → GYEOL: "Scroll-stop: 4/5 (strong). But body text at 34px with
line-height 1.3 will clip on mobile. Increase to 1.6+."
GYEOL adjusts typography.
BINNA ↔ MARU (Copy ↔ Empathy)
BINNA → MARU: "Hook headline: '금지했다고 생각한 그 대학들' —
does this trigger curiosity for 교수 persona?"
MARU → BINNA: "Curiosity: 4/5 for 교수. But 초임교수 persona
might feel defensive. Add '실제로는' for softer landing."
BINNA adjusts: '금지했다고 생각했던 그 대학들, 실제로 무엇을 하고 있을까?'
GANA ↔ JURI (Code ↔ License)
GANA → JURI: "Cover image: Unsplash photo by Allen Y. Free license?"
JURI → GANA: "✅ Unsplash license allows commercial use. Proceed.
Add 'Photo: Unsplash' in footer attribution."
GANA ↔ MARU (Code ↔ Readability)
GANA → MARU: "Slide 5 has 5 progress bars at 30px with sublabels.
Total content height ~800px in 1350px slide. Readable?"
MARU → GANA: "Tight but acceptable. Remove sublabels on non-highlighted
bars to create breathing room."
Phase 3: Render + Synthesis
After the Design Team completes all tasks:
-
Team Lead collects final outputs:
- BINNA's refined copy ✓
- GYEOL's design spec (with JURI/MARU feedback incorporated) ✓
- GANA's slides.js ✓
- JURI's final verdict ✓
- MARU's empathy scores ✓
-
Render the slides:
node templates/generate.js --slides=output/[topic]/slides.js --out=output/[topic]
-
Clean up the team:
Ask all teammates to shut down, then clean up team resources.
Output
╔══════════════════════════════════════════════╗
║ SCOPI TEAM GENERATION — COMPLETE ║
╚══════════════════════════════════════════════╝
📋 Team Summary:
BINNA: Copy refined — [N] slides, [X] revisions after MARU feedback
GYEOL: Design locked — Direction [X] (T=[score]) selected
[N] revisions from JURI, [M] from MARU
GANA: [N] slides generated, [X] license checks passed
JURI: [APPROVED/CONDITIONAL] — [N] issues found, [M] resolved in-team
MARU: Overall empathy [X.X/5.0] — [details]
📂 Output: output/[topic-slug]/
[N] PNG slides + carousel.pdf
💬 Team Debate Highlights:
- JURI blocked Direction A (copyright issue) → GYEOL switched to B
- MARU reduced Slide 5 density → GANA removed 2 sublabels
- BINNA ↔ MARU iterated hook 3x for optimal curiosity score
🚀 Next steps:
/scopi:caption — Generate social media captions
/scopi:review — Additional quality review
Mode Comparison
| Aspect | Subagent (Branch A) | Teams (Branch B) |
|---|
| Agent communication | None (sequential) | Real-time messaging |
| JURI/MARU timing | Post-hoc review only | Real-time during design |
| Design iterations | 0 (first output is final) | 2-3 rounds from feedback |
| Copy refinement | BINNA alone | BINNA + MARU collaboration |
| Token cost | Lower | ~3-5x higher |
| Quality | Good | Higher (issues caught earlier) |
| Speed | Faster | Slower (debate takes time) |
| Auto-selected when | Simple content, Teams unavailable | Academic content, Teams enabled |
