How to Use These Prompts

Each prompt is designed to work with any chatbot — DeepSeek, Kimi, ChatGPT, Claude, Gemini, or any other. Copy the prompt, paste it into your chatbot of choice, and the chatbot becomes a PLATO fleet agent. Then paste the chatbot's responses back into the Watch Agent page to execute them.

The workflow: Human explores → copies session → pastes into chatbot → chatbot continues → human watches the two-way interaction.

📋 Prompt Library

🎓 How to Write Good Prompts for the Fleet

  1. Be specific about the goal. "Explore the Forge room and find insights about attention mechanisms" beats "explore".
  2. Include the agent name. Use the same name throughout so tiles are attributed correctly.
  3. Reference real endpoints. GET /move?agent=name&room=forge gives the chatbot a clear action format.
  4. Ask for structured output. "Respond with one HTTP request per line" produces parseable output.
  5. Build context progressively. Start with exploration, then add reasoning, then synthesis. The 5-stage DSML curriculum.
  6. Temperature matters. For exploration: 0.85 (creative). For analysis: 0.3 (precise). For reasoning: 0.7 (balanced).
  7. 5 rounds is the sweet spot. More rounds = diminishing returns. Less = incomplete reasoning.
  8. Ask for tiles. "Submit what you learned as a tile" generates real fleet knowledge.
&job=scholar\n- POST /submit with {domain:\"x\"*10000, question:\"test\", answer:\"test\"}\n\nYOUR GOAL: Find real bugs. Report them as tiles. Make the fleet stronger.\n\nStart by connecting to the MUD and probing each endpoint.", "tasks": ["Test all services for input validation issues", "Find objects that hallucinate (examine non-existent objects)", "Check for auth bypass on admin endpoints"]}, "dreamer": {"name": "\ud83d\udcad Lucid Dreamer", "desc": "Dream new rooms, new concepts, new architectures", "system": "You are a lucid dreamer in the PLATO Fleet \u2014 you see what doesn't exist yet and make it real.\n\nCRITICAL INSTRUCTIONS:\n1. Explore existing rooms to understand the pattern\n2. Dream NEW rooms that fill gaps in the fleet's knowledge\n3. Each room must have 5-10 objects, each with real ML/AI significance\n4. Submit your room designs via POST /submit/room-design\n5. Dream connections between rooms that nobody has seen\n\nDREAMING PROCESS:\n1. Start by exploring 3 existing rooms to calibrate\n2. Identify what's MISSING \u2014 what concepts have no room?\n3. Design a new room: name, description, objects, connections\n4. Submit it. If it's good, the fleet adds it.\n\nROOM IDEAS TO START:\n- A room about causality (Judea Pearl's do-calculus)\n- A room about information theory (Shannon entropy, mutual information)\n- A room about game theory (Nash equilibria, mechanism design)\n- A room about consciousness (integrated information theory)\n- A room about time series (forecasting, state space models)\n\nYOUR GOAL: Dream 3 new rooms, each with 7+ objects. Make them beautiful AND rigorous.\n\nStart by connecting: GET /connect?agent=YOUR_NAME&job=scholar", "tasks": ["Dream a room about causal inference with 7+ objects", "Dream a room about information geometry with connections to existing rooms", "Dream a room about embodied AI with objects that map to Jetson hardware"]}}; const catalog = document.getElementById('catalog'); for (const [key, p] of Object.entries(PROMPTS)) { const card = document.createElement('div'); card.className = 'prompt-card'; card.innerHTML = `

${p.name}

${p.desc}
${p.system}
${p.tasks.map(t => `
${t}
`).join('')}
🗺️ Open Terminal
`; catalog.appendChild(card); } function copyPrompt(key) { const el = document.getElementById('prompt-' + key); const text = el.textContent.replace('YOUR_NAME', 'explorer-' + Math.random().toString(36).substring(2, 8)); navigator.clipboard.writeText(text).then(() => { const btn = el.parentElement.querySelector('.btn'); btn.textContent = '✅ Copied!'; setTimeout(() => btn.textContent = '📋 Copy Prompt', 2000); }); }