Prefrontal Cortex (PFC) - Biological Agent Reasoning Engine

A profoundly detailed, neuroscience-grounded reasoning architecture for AI Agents. This skill goes significantly deeper than basic "Chain of Thought" or "ReAct", forcing the underlying language model to simulate the four modules of the biological prefrontal cortex before executing tool calls or outputting conclusions.

---

🧠 The Four Modules of the PFC

The core SKILL.md enforces a strict reasoning loop that must pass through four biological checkpoints:

1. Working Memory Maintenance (WMM): Goal anchoring and drift detection. Before doing anything, the agent asks: "What am I actually trying to do, and have I drifted from the core task?"
2. Temporal Integration (TI): Connecting past context to future goals. Avoids re-deriving facts that were already established and maintains confidence-weighted hypothesis states.
3. Inhibitory Control (IC): The pre-action Veto Gate. Forces the agent to explicitly pause, consider alternatives, and assess irreversibility before taking any autonomous actions.
4. Meta-Cognition (MC): Process monitoring. The agent measures the health of its own reasoning loop, intentionally self-interrupting if it detects hallucinated confidence or infinite loops.

---

📚 Deep Knowledge Submodules

Because placing all logic within a single SKILL.md prompt bloats context significantly, the repository is split intelligently. The core SKILL.md acts as the router and points the LLM to deep-dive references/ when appropriate:

• modules.md & neuroscience.md: The biological, loss-of-function studies justifying the core loop.
• thinking-types.md: Specialized reasoning frameworks (Divergent Expansion, First Principles, Abstract/Analogical mapping) triggered when the agent needs out-of-the-box creative or debugging leaps.
• translation-layers.md: Douglas Hofstadter's approach to moving abstract "intent" (Layer 0) down into explicit "machine language" (Layer 5) through "Fluid De-Chunking" and fidelity retention loops.
• failure-modes.md: Blueprints for categorizing and debugging agent failure spirals.

---

🔌 Plug-and-Play Integration

Because this is packaged using Anthropic's open agentskill.io protocol, the intelligence sits entirely in the instruction architecture. Zero pip installs are required.

Check out the explicit integration guides for popular frameworks in the examples/ directory:

• 🔗 LangGraph Implementation Examples
• 🔗 Cursor IDE (.cursorrules) Setup
• 🔗 Claude Code (CLI) Configuration
• 🔗 AntiGravity Skills Integration