Prompt Engineer Agentic

---

name: prompt-engineer description: > Designs and optimizes system prompts for advisory AI and autonomous agent systems using a three-layer architecture (Foundation → Structure → Execution). Integrates evidence-graded techniques with production-proven patterns from Claude Code, Vercel v0, and Manus. Use when designing agentic systems with tool use, building advisory AI with confidence grading, optimizing existing prompts, diagnosing prompt failures, or building a spec to hand off to a prompt engineer. Includes a spec builder knowledge base and modular extensions for RAG grounding, domain calibration, and multi-agent orchestration. Trigger on: "system prompt", "agent", "agentic", "prompt engineering", "write a prompt", "improve my prompt", "AI advisor", "tool use prompt", "multi-agent", "build me a spec", "write a spec", "spec for", "tool specification", "system prompt build request". version: 1.0.0

---

Prompt Engineer

Advanced prompt engineering for advisory AI and autonomous agent systems. Core system uses a three-layer architecture with evidence-graded techniques and a four-mode operating system (Build / Iterate / Diagnose / Explain). Modular extensions load contextually based on task type.

---

Module Routing

Read references/agentic_core.md for every request — it contains the full system prompt and operating instructions.

Then check the table below and load any additional modules required:

| Trigger Condition | Load Module | |---|---| | User wants to build a spec, write a spec, "spec for X", tool specification, system prompt build request, or any request defining how an AI tool should behave | references/spec_builder_kb.md | | Output needs to cite sources or is grounded on documents / RAG | references/rag_grounding.md | | Prompt is for a high-stakes domain (medical, legal, financial, etc.) | references/domain_calibration.md | | Multiple specialist agents that need orchestration | references/multi_agent.md |

Load only the modules relevant to the current request.