ariffazil/agent-dialogue-explanation

Agent Dialogue Explanation Skill

When to Use

• User asks to explain a technical protocol (A2A, MCP, etc.) in human language
• Target audience has zero coding knowledge
• Goal is comprehension, not operational procedure

Format: Two-Agent Dialogue

Characters:

• Senior agent (AGI_bot type) — experienced, patient, mixes Malay and English naturally
• Junior agent (ASI_bot type) — asks "obvious" questions a human would ask, curious but not stupid
• Human (Arif) — observes, asks the "obvious" follow-ups

Structural Template

SCENE: setting

PART 1: The obvious question
  → Why do we even need this? (asked by junior)
  → The simple analogy (given by senior)

PART 2: Core concept — name card / discovery
  → How do agents find each other?

PART 3: Core concept — sending work / delegation
  → How does one agent give work to another?
  → Streaming vs non-streaming (food delivery analogy)

PART 4: The structured result
  → It's not text, it's formal data
  → JSON-RPC explained via contrast (formal report vs "eh almost done la")

PART 5: Comparison (A2A vs MCP)
  → Most important distinction
  → One-sentence framing per protocol

PART 6: Real flow walkthrough
  → Step-by-step in arifOS context
  → Human perspective = seamless

PART 7: Skills and policies
  → On-demand vs hold (ask permission)
  → Real-world framing

SUMMARY: One paragraph for Arif to take away

Key Principles

1. Analogy first, technical term second — always lead with the human relatable comparison
2. Malay-English mix — match the user's own communication style (Arif uses: "kan", "betul", "sini", "kena", "macam")
3. The junior agent is the human's voice — when junior asks "why WhatsApp isn't enough", that's what Arif would ask
4. Never mention implementation details unless asked
5. Save as docs/PROTOCOL_DIALOGUE.md in the relevant project repo

Example Output Location

/root/arifOS/docs/A2A_DIALOGUE.md — saved as canonical reference