Multi-agent orchestration: workflow execution, triad composition, routing, parallelism, subagent monitoring, error recovery, and learning loops. Topics: continuous-learning, error-recovery-automation, intelligent-routing, multi-model-routing, parallel-workflow, socratic-debate, subagent-monitor, task-automation, triad-composition, workflow-orchestration.
Router skill: dispatches multi-agent orchestration work to exactly one playbook. [DOC]
Any request to compose/route/run/recover/monitor a multi-agent workflow under PRISTINO. [DOC] NOT for single-agent tasks, content generation, or domain skills — route those to their own skill. [INFERENCE]
topic (one of the 10 enum values; infer from request, ask only if two routes tie) + depth. [CONFIG][CODE]/[CONFIG]/[DOC]/[INFERENCE]/[ASSUMPTION]) — never mix tag families. [DOC]topic from the request. Read EXACTLY ONE playbook from routes:. Never load the cluster. [CONFIG]depth=deep → apply exhaustively, verify at each step; quick → essentials only. Default quick. [CONFIG]triad-composition; which model per task → multi-model-routing. [DOC]intelligent-routing; end-to-end run → workflow-orchestration. [DOC]parallel-workflow; live subagent status → subagent-monitor. [DOC]socratic-debate; harvesting its insights → continuous-learning. [DOC]error-recovery-automation; schedule/trigger work → task-automation. [DOC]Done only if: one playbook loaded (not many); spine completed; gates passed; tags single-family + EN/ES consistent. [DOC]
Walk assets/routing-checklist.md before emitting; Guardian scores with assets/quality-rubric.json. [CONFIG]
topic when two routes genuinely tie — ask instead of mis-routing. [ASSUMPTION]If mid-run the request no longer fits the chosen topic, stop, re-resolve against the
disambiguation table, and switch playbooks — do not stretch the wrong one. [INFERENCE]
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.