Agent Core Suite¶
Core orchestration, advanced reasoning, and context engineering. Uses the Hub Skill architecture with 4 hubs routing to 13 sub-skills. Optimized for Claude Opus 4.7 with adaptive thinking and Agent Team support.
Version: 3.5.2 | 3 Agents | 2 Registered Commands | 4 Hubs → 13 Sub-skills | 12 Hook Events
Agents¶
Agent: orchestrator
Multi-agent orchestrator specializing in workflow coordination, agent team assembly, and task allocation.
Model: opus
Version: 3.5.2
Agent: reasoning-engine
Expert in advanced reasoning, prompt design, and cognitive tasks. Masters Chain-of-Thought and structured frameworks.
Model: opus
Version: 3.5.2
Agent: context-specialist
Elite AI context engineering specialist mastering dynamic context management, vector databases, and memory systems.
Model: opus
Version: 3.5.2
Registered Commands¶
Command: /ultra-think
Comprehensive analysis with full reasoning framework execution.
Command: /team-assemble
Generate ready-to-use agent team configurations from pre-built templates.
Skill-Invoked Commands¶
These commands are triggered by skills, not directly by users:
Command: agent-build
Unified AI agent creation, optimization, and prompt engineering.
Command: ai-assistant
Build production-ready AI assistants with NLU and intelligent response generation.
Command: docs-lookup
Query library documentation using Context7 MCP for up-to-date API references.
Command: reflection
AI reasoning analysis, session retrospectives, and research optimization.
Hub Skills¶
Skills use a hub architecture: 4 hub skills route to 13 specialized sub-skills.
Hub: agent-systems¶
Multi-agent coordination, performance optimization, evaluation, and tool use patterns.
agent-evaluation— Evaluate AI agent performance through systematic testing and benchmarkingagent-performance-optimization— Monitor, cache, and load-balance agent systems for productionmulti-agent-coordination— Workflow orchestration, task allocation, and inter-agent communicationtool-use-patterns— Tool selection, chaining, error handling, and result synthesis
Hub: reasoning-and-memory¶
Reasoning frameworks, reflection, knowledge graphs, memory systems, and closed-loop self-improvement.
reasoning-frameworks— First Principles, RCA, Decision Analysis, Systems Thinking, OODA Loopreflection-framework— Meta-cognitive analysis, bias detection, and session reflectionknowledge-graph-patterns— Knowledge graphs for structured retrieval and semantic reasoningmemory-system-patterns— Persistent memory systems with vector stores and context managementself-improving-agents— Reflection-refine-validate loops, self-consistency ensembles, DSPy/TextGrad prompt optimization, evolutionary prompt search, constitutional self-critique (new in v3.1.4)
Hub: llm-engineering¶
Intent clarification, prompt engineering, LLM application patterns, MCP integration, and safety.
thinkfirst— Interview-first workflow that clarifies vague intent through a Seven Dimensions framework before any prompt is drafted (new in v3.1.3)llm-application-patterns— Prompt engineering principles (CoT, few-shot), RAG design, evaluationmcp-integration— MCP server configuration, tool naming conventions, and cross-tool workflowsprompt-engineering-patterns— Advanced prompting with chain-of-thought and production templatessafety-guardrails— Content filtering, output validation, and responsible AI practices
Hooks¶
12 hook events with Python script implementations:
SessionStart— Session initializationSessionEnd— Session teardownPreToolUse— Before tool executionPostToolUse— After tool executionPreCompact— Before context compactionPostCompact— After context compactionSubagentStart— When a subagent startsSubagentStop— When a subagent completesPermissionDenied— When a tool call is deniedTaskCreated— When a task is createdTaskCompleted— When a task finishesStopFailure— On agent stop failure
(PreSubagentUse, ExecutionError, PermissionPrompt, ContextOverflow, and CostThreshold handlers were removed in v3.4.0 — not supported by the CC v2.1.113 CLI event schema.)