AI/LLM system architecture: software arch, pipelines, conops, design patterns, audit, implementation. Topics: ai-conops, ai-design-patterns, ai-pipeline-architecture, ai-software-architecture, audit, chatbot-design, embedding-strategy, fine-tuning-prep, implementation, prompt-engineering, rag-patterns, structured-output, voice-interface.
Router skill: resolve topic from the request, then Read EXACTLY ONE playbook
from routes:. The playbook does the work; this file only routes. [DOC]
User asks to design, audit, or implement an AI/LLM system component — software architecture, pipelines, ConOps, design patterns, RAG, prompts, embeddings, fine-tuning prep, structured output, chatbot/voice interface. If the request is not LLM/AI-system shaped, do not route here. [INFERENCIA]
topic (one enum value, required), depth (quick|deep, default quick).topic: infer from the request; ask ONLY if two enum values fit
equally. Map intent, not keywords (e.g. "make my prompt reliable" →
prompt-engineering; "ground answers in our docs" → rag-patterns). [INFERENCIA]topic; if none fits, say so and stop — do not force-fit. [SUPUESTO]routes: playbook. Read NOTHING else from the cluster. [DOC]quick → essentials only; deep → apply the playbook exhaustively, verifying
at each step. Honor the playbook's own gates over these defaults. [INFERENCIA]Spine: Discover → Analyze → Execute → Validate.
Quality gates: constitution v6.0.0 (enforcement), evidence tags, script-first rule. [CONFIG]
Deterministic routing aids: assets/ (routing-rubric.json, routing-checklist.md). [CONFIG]
topic resolved and present in the enum. [DOC][DOC] [CONFIG] [CÓDIGO] [INFERENCIA] [SUPUESTO]). [CONFIG]topic when the request already makes it obvious. [SUPUESTO][CÓDIGO]/[CONFIG] you cannot point to in-repo — downgrade to [SUPUESTO]. [CONFIG]If the playbook doesn't fit the ask, stop, re-resolve topic, Read the correct
one — never patch with a second. If no enum fits, report the gap. [SUPUESTO]
JaviMontano/jm-adk-beta
fdad39c
If you maintain this skill, you can claim it as your own. Once claimed, you can manage eval scenarios, bundle related skills, attach documentation or rules, and ensure cross-agent compatibility.