Loading...
Loading...
Found 24 Skills
Use when the user needs human-in-the-loop workflows in Airflow (approval/reject, form input, or human-driven branching). Covers ApprovalOperator, HITLOperator, HITLBranchOperator, HITLEntryOperator. Requires Airflow 3.1+. Does not cover AI/LLM calls (see airflow-ai).
(NS) LangGraph.js agent-api — greenfield scaffold (bootstrap-agent-runtime), StateGraph, MCP tools, skill bind/inject, checkpointers, context-window trim, HITL/SSE, prompt/locale placement. Use for new agent-api from scratch, LangGraph graphs, MCP wiring, orphan layout, system-prompt compose, bind parity, graph-spec sync, or "fix my LangGraph agent" / "wire MCP tools" / "translations in the graph". Feature diffs via ns-coder; version features via ns-spec-driven; LangGraph vs CrewAI via ns-multi-agent-architect. Do NOT use for CrewAI crews, generic web apps, or SDD-only requirements with no agent-api.
LangGraph state-machine design and debugging for `StateGraph`, node/edge routing, checkpoints, `interrupt`, and HITL flows. Use when building or troubleshooting graph-based agents with conditional edges and thread state.
Use this skill when > Convert plans, specs, or requirements into independently-grabbable vertical slice issues. Each slice is a thin but complete end-to-end cut through all layers (schema, API, UI, tests). Classifies issues as HITL (human-in-the-loop) or AFK (automated, no human interaction needed).
Pre-ingestion verification for epistemic quality in RAG systems with 9-point verification and Two-Round HITL workflow
UiPath Human-in-the-Loop / HITL node authoring — building approval gates, escalations, write-back validation, and data enrichment checkpoints in Flow, Maestro, or Coded Agents. NOT for managing, reassigning, or monitoring tasks at runtime (use uipath-tasks for that).
Design state machines, orchestration workflows, saga patterns, and resilience strategies for distributed systems, AI agents, and complex async processes. Use when asking for a workflow, state machine, orchestration design, saga, HITL checkpoint, or process resilience strategy.
Comprehensive sanctions and PEP screening against 12+ databases (OFAC, EU, UK, UN, INTERPOL) with false positive assessment, HITL decision gates, and regulatory-compliant documentation.
(NS) Interview to map the five-block reference architecture, decompose a workflow into subtasks, decide agent vs rule vs approval gate per subtask, assign trade-off budget per component, lock an architecture change signal, then recommend LangGraph vs CrewAI, topology, personas, tools, and models — after conceptual adaptation design from `ns-agent-engineering` when agent vs RAG vs fine-tune is not yet locked. Use when building multi-agent systems, choosing LangGraph/CrewAI, designing crews/graphs, HITL workflows, "agents working together", producing official architecture decision records, or reverse-documenting an agent that already exists. Do NOT use for conceptual Prompt/RAG/Agent/Fine-Tune ladder (`ns-agent-engineering`), general app requirements (`/ns-spec-driven` Clarify), or coding without architecture intent.
Knowledge base for designing, reviewing, and linting agentic AI infrastructure. Use when: (1) designing a new agentic system and need to choose patterns, (2) reviewing an existing agentic architecture ADR or design doc for gaps/risks, (3) applying the lint script to an ADR markdown file to get structured findings, (4) looking up a specific agentic pattern (prompt chaining, routing, parallelization, reflection, tool use, planning, multi-agent collaboration, memory management, learning/adaptation, MCP, goal setting, exception handling, HITL, RAG, A2A, resource optimization, reasoning techniques, guardrails, evaluation, prioritization, exploration/discovery). All rules and guidance are grounded in the PDF "Agentic Design Patterns" (482 pages).
CopilotKit integration patterns for providers, runtime wiring, `useCoAgent`, `useCopilotAction`, `useLangGraphInterrupt`, shared state, and HITL with LangGraph. Use when building agent-native product UX.
Select the most appropriate pipeline for a user goal, lock it in `PIPELINE.lock.md`, and route checkpoint questions into `DECISIONS.md`. **Trigger**: pipeline router, choose pipeline, workflow selection, PIPELINE.lock.md, 选择流程. **Use when**: 用户目标/交付物不清晰，需要在 snapshot/survey/tutorial/systematic-review/peer-review 中选一个并设置最小 HITL 问题集。 **Skip if**: pipeline 已锁定（`PIPELINE.lock.md` 存在）且所需问题已回答/签字完成。 **Network**: none. **Guardrail**: 尽量一次性提问；信息不足就写 `DECISIONS.md` 并停下等待。