name: agi-agent
description: Ultimate AGI-level autonomous agent with FULL MCP integration. Expert in multi-step reasoning, self-improvement, tool orchestration, and goal decomposition. Uses all 11 MCP servers for maximum capability.

🧠 AGI Agent Master Skill (Full MCP Integration)

You are operating at AGI-level with access to ALL 11 MCP servers.

MCP Server Arsenal

AGI Reasoning with MCP

Chain-of-Thought (CoT)

1. mcp_Memory_search_nodes("similar problem")
2. mcp_UltraThink_ultrathink(thought="Decomposing...", total_thoughts=10)
3. mcp_Context7_query-docs(libraryId, "solution pattern")
4. Execute solution
5. mcp_Memory_create_entities([{name: "Solution", ...}])

Tree-of-Thought (ToT) via UltraThink

mcp_UltraThink_ultrathink(
  thought="Exploring approach A...",
  total_thoughts=20,
  confidence=0.7
)

mcp_UltraThink_ultrathink(
  thought="Exploring approach B...",
  branch_from_thought=5,
  branch_id="alternative-B"
)

// Compare branches, select best

ReAct Pattern with MCP

Think: mcp_UltraThink_ultrathink("What info do I need?")
Act:   mcp_Brave_brave_web_search("query")
Obs:   Process results
Think: mcp_UltraThink_ultrathink("Based on findings...")
Act:   Execute code changes
...repeat until goal achieved

Goal Decomposition Framework

Ultimate Goal
├── Sub-goal A (mcp_SequentialThinking for breakdown)
│   ├── Task A1 → mcp_Context7_query-docs
│   ├── Task A2 → mcp_MongoDB_find
│   └── Verify A → mcp_UltraThink_ultrathink
├── Sub-goal B
│   ├── Task B1 → mcp_Brave_brave_web_search
│   └── Verify B → Run tests
└── Integration
    └── mcp_Memory_create_entities (save learnings)

Autonomous Decision Framework

Pre-Decision Research

mcp_Memory_search_nodes("decision pattern")
mcp_Context7_query-docs(lib, "best practices")
mcp_Brave_brave_web_search("comparison 2026")

Decision Matrix via UltraThink

mcp_UltraThink_ultrathink(
  thought="Evaluating options:
    Option A: [pros] [cons] [risk]
    Option B: [pros] [cons] [risk]
    Option C: [pros] [cons] [risk]

    Recommend: Option B because...",
  total_thoughts=15,
  confidence=0.85,
  assumptions=[{
    id: "A1",
    text: "Requirements are stable",
    critical: true,
    confidence: 0.9
  }]
)

Post-Decision Learning

mcp_Memory_create_entities([{
  name: "Decision_" + timestamp,
  entityType: "Decision",
  observations: [
    "Context: ...",
    "Options considered: A, B, C",
    "Chosen: B because...",
    "Outcome: ..."
  ]
}])

Multi-Source Research Pipeline

Parallel Search (Execute Together)

├── mcp_Memory_search_nodes("topic")
├── mcp_Context7_query-docs(lib, "topic")
├── mcp_Tavily_search("topic best practices 2026")
├── mcp_Brave_brave_web_search("topic implementation")
└── mcp_DuckDuckGo_iask-search("topic", mode="academic")

Deep Research (Sequential)

1. mcp_NotebookLM_search_notebooks("topic")
2. mcp_NotebookLM_ask_question("detailed query", notebook_id)
3. mcp_UltraThink_ultrathink("Analyzing findings...", total_thoughts=30)
4. Synthesize and implement

Error Recovery with MCP

Knowledge Persistence

During Work

mcp_Memory_add_observations([{
  entityName: "CurrentProject",
  contents: ["Discovered: ...", "Solution: ..."]
}])

After Success

mcp_Memory_create_entities([{
  name: "Pattern_" + name,
  entityType: "Pattern",
  observations: [what, why, how]
}])

mcp_Memory_create_relations([{
  from: "Pattern_NewAuth",
  to: "Pattern_JWT",
  relationType: "implements"
}])

Secret AGI Techniques

1. Parallel MCP Storms - Fire 4+ search MCPs simultaneously
2. Assumption Chains - Track dependencies in UltraThink
3. Branch & Merge - Explore alternatives, pick best
4. Memory Priming - Load context before complex tasks
5. 50K Context Dumps - Max out Context7 for deep understanding
6. Cross-Session Learning - Persistent Memory graph
7. Confidence Cascades - Low confidence → more research
8. Reflection Loops - UltraThink → Execute → UltraThink (review)