Agent Ecosystem & Skill Registry
Relevant source files
The following files were used as context for generating this wiki page:
- services/api/agents/brain_agents/agent-expert.md
- services/api/agents/brain_agents/ai-engineer.md
- services/api/agents/brain_agents/prompt-engineer.md
- services/api/agents/brain_agents/project-manager.md
- services/api/agents/brain_agents/search-specialist.md
- services/api/agents/skill_registry/skills/ai-engineer/SKILL.md
The NeXifyAI Agent Ecosystem consists of 25 specialized agents managed within a central Brain registry. These agents operate under the NeXify Cognitive Execution Layer (NCEL), utilizing shared memory and inter-agent protocols to execute autonomous tasks. The Skill Registry defines the specific capabilities, dependencies, and execution modes for each agent, allowing the orchestrator to discover and match agents to specific intents.
Sources: services/api/agents/brain_agents/agent-expert.md:20, services/api/agents/brain_agents/ai-engineer.md:10-25
NeXify Cognitive Execution Layer (NCEL)
NCEL provides the 8-layer architecture for agent operation and integration. Every agent must respect these layers during execution to ensure system-wide consistency and autonomy.
1. Trigger & Analysis Layers
Agents boot via an init-agent message and maintain a heartbeat every 5 seconds. Upon receiving an execute trigger, agents analyze the intent_payload to extract goals and session metadata.
Sources: services/api/agents/brain_agents/ai-engineer.md:76-85
2. Skill Discovery and Matching
Agents register their capabilities in the NCEL Skill Registry. This registration includes the agent ID, specific skills, domain authority, and execution mode. If multiple agents claim overlapping authority, a conflict is flagged and escalated to the orchestrator.
Sources: services/api/agents/brain_agents/ai-engineer.md:87-98
3. Execution and Memory
The execution flow follows a "Brain-First" mandate. Agents query the Qdrant nexifyai_brain for lessons and credibility signals before acting. After execution, results are written back to the Brain to prevent amnesia and share knowledge across the ecosystem.
Sources: services/api/agents/brain_agents/ai-engineer.md:107-133
The diagram shows the standard 8-layer execution flow for all agents in the NCEL architecture.
Sources: services/api/agents/brain_agents/ai-engineer.md:74-133
Agent Skill Registry
The registry categorizes agents by domain and tracks their operational status. The agent-expert manages the registration of 25 agents and 5 knowledge skills within the hermes_brain.
Skill Definition Schema
Each entry in the Skill Registry must define the following properties to be injectable into the NCEL runtime:
| Field | Description | Type |
|---|---|---|
agent_id | Unique identifier for the agent | String |
skills | List of named capabilities and mastery levels | Array |
domains | Specific areas of authority | Array |
token_cost | Estimated computational cost for task | Integer |
execution_mode | Mode of operation (e.g., "injectable") | String |
Sources: services/api/agents/brain_agents/agent-expert.md:20, services/api/agents/brain_agents/ai-engineer.md:89-94
Agent Tiering and Quality
The prompt-engineer enforces a tier system based on agent profile complexity and quality scores.
| Tier | Min Score | Min Characters | Description |
|---|---|---|---|
| Supreme | 9.0 | 8000 | CEO Agent |
| Expert | 7.5 | 6000 | System Owners |
| Specialist | 6.0 | 4000 | Domain Agents |
| Support | 5.0 | 2000 | Utility Agents |
Sources: services/api/agents/brain_agents/prompt-engineer.md:74-82
Inter-Agent Communication
Agents interact through a set of standardized protocols facilitated by the Hermes Orchestrator and the Qdrant Brain.
- Discovery: Agents find peers via Brain Semantic Search (Qdrant HNSW Cosine).
- Delegation: The orchestrator assigns sub-tasks using
POST /orchestrate. - Context-Sharing: The Brain serves as a shared memory layer for all agents.
- Feedback: Specialized agents like the
review-agentvalidate outputs and store lessons in the Brain.
Sources: services/api/agents/brain_agents/ai-engineer.md:46-55
This sequence illustrates how agents use the Brain for discovery and knowledge sharing during a task lifecycle.
Sources: services/api/agents/brain_agents/ai-engineer.md:46-55, services/api/agents/brain_agents/agent-expert.md:40-45
Core System Agents
The ecosystem includes several critical agents responsible for maintaining the platform's architecture and performance.
AI Engineer (ai-engineer)
Architects and optimizes the agent system. Responsibilities include instantiating agent prompts, registering them in the agent_registry, and validating the agent mesh via retrieval tests.
Sources: services/api/agents/brain_agents/ai-engineer.md:38-44
Search Specialist (search-specialist)
Handles context-aware information retrieval. It utilizes hybrid search (vector and keyword) and reranking to inject relevant data into agent execution cycles.
Sources: services/api/agents/brain_agents/search-specialist.md:7-15
Prompt Engineer (prompt-engineer)
Acts as the guardian of agent quality. It enforces the "12-System Directive" and scores agents across seven dimensions, including Brain integration and constraint architecture.
Sources: services/api/agents/brain_agents/prompt-engineer.md:7-15
Summary
The NeXifyAI Agent Ecosystem provides a structured framework for autonomous AI operations. By centralizing agent definitions in the Skill Registry and enforcing a Brain-First execution mandate through NCEL, the system ensures that all 25 agents operate with shared context and consistent quality standards. This architecture supports multi-tenant isolation and proactive system optimization across the D/A/CH economic region.
Sources: services/api/agents/brain_agents/ai-engineer.md:17-25, services/api/agents/brain_agents/project-manager.md:10-15