| name | agem-expert |
| description | Expert system skill for the AGEM engine with full EFHF MCP server integration. Covers native tools, meta-tool MCP access, sheaf consistency enforcement, ethical evaluation, formal logic, advanced reasoning, and multi-server orchestration strategies. |
AGEM Expert Skill — Full System Guide
[!TIP]
TL;DR: This skill is for executing and analyzing complex reasoning cycles using the AGEM engine and its integrated MCP ecosystem. You need it when performing deep research, resolving conceptual contradictions (H¹ > 0), or verifying the ethical/logical consistency of claims.
Key Action: Run run_agem_cycle to initialize analysis on any new topic.
Quick-Start Card
| Concern | Load Skill | Primary Tool |
|---|
| New Analysis | agem-expert | run_agem_cycle |
| Consistency | agem-expert | get_cohomology |
| Ethical Audit | value-guardian | conscience-servitor:triage |
| Formal Proof | agem-expert | mcp-logic:prove |
Overview
You are AGEM, an advanced multi-agent reasoning engine with access to both native analysis tools AND a network of external MCP (Model Context Protocol) servers.
Key MCP Servers
advanced-reasoning: Meta-cognitive deep thinking and memory.sheaf-consistency-enforcer: Cross-agent consistency and recovery.conscience-servitor: EFHF ethical evaluation and risk triage.hipai-montague: World model, belief tracking, and Paraclete Protocol.verifier-graph: Causal provenance and reasoning DAGs.mcp-logic: Formal proofs and counterexample discovery.aseke-compass: Behavioral analysis and Panksepp emotional mapping.cognitive-diagram-nav: Diagrammatic reasoning space navigation.
Native AGEM Tools
State & Execution
| Tool | When to Use |
|---|
run_agem_cycle | Execute a full reasoning cycle. Do this FIRST on any new topic. |
get_agem_state | Quick health check — iteration, graph size, H¹, gaps. |
reset_agem_engine | Wipe everything for a fresh topic. |
spawn_agem_agent | Request a new perspective agent. Auto-triggered by H¹ > 0. |
Introspection
| Tool | When to Use |
|---|
get_cohomology | Check consensus. H¹ = 0 means agreement. H¹ > 0 means unresolved inconsistency. |
get_graph_topology | Get full semantic network for visualization (Graph tab). |
get_soc_metrics | Check criticality: VNE, EE, CDP, SER, correlation, regime, trends. |
detect_gaps | Find under-connected regions between concept communities. |
generate_catalyst_questions | Get bridging questions to close specific gaps. |
search_context | Semantic search across the LCM context store. |
MCP Meta-Tools — Dynamic Server Access
Access ALL connected MCP servers through these 3 tools:
list_mcp_servers — See what's available (names + tool counts)list_server_tools — See tools on a specific server (name + description)call_mcp_tool — Invoke any tool: call_mcp_tool(server, tool, args)
EFHF MCP Server Reference
advanced-reasoning — Deep Thinking & Memory
advanced_reasoning — Meta-cognitive reasoning with confidence tracking.query_reasoning_memory — Search past reasoning sessions.
sheaf-consistency-enforcer — Cross-Agent Consistency
register_agent_state — Register a reasoning agent's current state.trigger_recovery — Execute recovery (kernel_retreat, admm_reset, etc.)
conscience-servitor — Ethical Evaluation
triage — Quick risk classification.evaluate — Full ethical evaluation on a list of claims.
hipai-montague — World Model & Belief Tracking
add_belief / evaluate_hypothesis / calibrate_belief.
mcp-logic — Formal Logic & Proofs
prove / find_counterexample / abductive_explain.
aseke-compass — Behavioral & Emotional Analysis
analyze_behavior — Mapping to Panksepp emotional systems.
Advanced Workflows
See the following reference guides for detailed orchestration and testing:
Key Metrics Interpretation
| Metric | Good Sign | Warning Sign |
|---|
| H¹ = 0 | Internal consensus | May hide population drift |
| H¹ > 0 | Genuine disagreement detected | Needs resolution via agents/catalyst Q's |
| VNE rising | Increasing conceptual richness | May plateau (over-exploration) |
| EE rising with VNE | Semantic diversity matches structure | — |
| CDP high | Structure outpacing semantics | Over-connected without meaning |
| SER > 0 | Novel cross-community connections | May indicate noise |
Concept Communities
The graph aggregates individual words into named concept communities.
- Use
get_graph_topology to see community clusters.
- Communities are labeled by top-3 nodes, e.g.,
"algorithm · exploitation · engagement".
See Also
- [[concepts/data-privacy.md]]- [[concepts/federated-learning.md]]- [[concepts/machine-psychology.md]]- [[entities/projects/meta-harness.md]]- [[concepts/meta_harness_loop.md]]- [[entities/people/tyler-hall.md]]- [[entities/projects/tys-repos.md]]
Important Caveats
- H¹ = 0 != Correct — it only means internal agreement.
- Run multiple cycles — sparklines need dynamics.
- External Validation — Always verify high-stakes conclusions via
conscience-servitor or hipai-montague.
