discover

name	discover
description	Trigger AI Discovery chain for exploring and optimizing prompts, skills, and agents through multi-agent debate
trigger	When user says "discover", "explore possibilities", "optimize prompts", "improve agents", "brainstorm with agents"

AI Discovery Chain

Launch multi-agent exploration and debate to discover novel solutions and optimize AI artifacts.

Usage

/discover <topic>
/discover optimize the marketing template for viral content
/discover improve the explorer agent's creativity
/discover find better patterns for error handling

What It Does

1. Phase 0: Intent Discovery — Clarifies what you're trying to achieve
2. Phase 1: Parallel Exploration — Multiple agents explore from different angles
3. Phase 2: Forum Debate — Agents challenge and stress-test ideas
4. Phase 3: Synthesis — Moderator extracts actionable insights
5. Phase 4: Optimization — Apply improvements to prompts/skills/agents
6. Phase 5: Validation — Verify changes work correctly
7. Phase 6: Report — Document discoveries and changes

Chain Flow

intent-discovery → parallel-exploration → forum-debate → synthesis
                           ↑                               ↓
                           └──── (max 2 rounds) ←──────────┘
                                                           ↓
                                              optimization → validation → report

Execution

When user invokes /discover:

1. Parse the topic from arguments
2. Spawn intent-discoverer agent with the topic to clarify intent and define exploration directive
3. Wait for intent analysis — this shapes all subsequent exploration
4. The chain continues automatically via hooks

Spawn Command

Agent(
  name: "intent-analyst",
  subagent_type: "intent-discoverer",
  prompt: "Analyze intent for: {user_topic}. Define exploration directive, success criteria, and what to avoid."
)

Options

• /discover <topic> — Full chain with intent discovery
• /discover fast <topic> — Skip intent discovery, jump to exploration
• /discover debate <ideas> — Skip exploration, go straight to debate on provided ideas

Output

The chain produces:

• Discovery Report — Summary of exploration, debates, and conclusions
• Optimized Artifacts — Improved agent/skill/prompt files (if optimization phase runs)
• Seeds for Future — Unresolved questions for next exploration cycle

Examples

/discover how to make the critic agent more constructive
/discover ways to reduce token usage in the development chain
/discover fast viral marketing hooks for tech products
/discover debate "use websockets vs SSE for real-time updates"