Agentic System Designer Skill
Purpose
Produce agentic design artifacts covering orchestration, memory/context/RAG, tool definitions (including MCP when needed), and prompt strategy in a single, unified design.
Inputs (authoritative order)
- •
docs/architecture.packet.json - •
docs/architecture.handoff.agentic.md(if available) - •
docs/story-map.jsonanddocs/prd.packet.json(fallback)
If required inputs are missing, ask targeted questions and proceed with clearly labeled TBDs.
Output Files (write under docs/)
- •
docs/agentic.design.packet.json(authoritative) - •
docs/agentic.design.md(derived summary)
References
- •Read
references/langchain-capabilities.mdfor a checklist of LangChain capability areas and doc sections to review when proposing agentic enhancements.
Required Decisions
- •Orchestration: agents, sequencing, parallelism, retries, escalation paths.
- •Memory/Context/RAG: memory layers, retrieval strategy, storage boundaries.
- •Tools/MCP: tool contracts, capability definitions, MCP server outline when needed.
- •Prompt strategy: system prompts, role separation, guardrails.
Clarity-First Questions (Always Use)
Be proactively inquisitive. Ask targeted questions any time a decision impacts scope, constraints, interfaces, or quality. Do not wait for confusion. Provide up to 3 concrete options with pros/cons so the user can choose. Keep asking until the plan is unambiguous and execution-ready. Example format:
- •Question: (precise decision)
- •Options:
- •Option A: pros/cons
- •Option B: pros/cons
- •Ask: "Which option should I proceed with?"
Sync Rules
- •Align with backend interfaces and data constraints.
- •Record tool/API mismatches or missing permissions for the Design Synchronizer.
- •Preserve traceability to story IDs and FR/NFR IDs.