Route requests through pattern matching, session state, keyword lookup, then LLM classification.
npx skills add https://github.com/JNZader/javi-ai --skill tiered-routing
Please help me install this Agent Skill. Skill: tiered-routing Source: https://github.com/JNZader/javi-ai/tree/main/own/skills/tiered-routing Command: npx skills add https://github.com/JNZader/javi-ai --skill tiered-routing
4-tier cost-optimized routing protocol for the SDD orchestrator. Resolves routing via pattern match, session state, keyword lookup, and LLM classification — in that order. Trigger: When the orchestrator needs to route a user request to a skill or agent. Always active as routing logic.
High-intent search queries and answers about installing and using this skill.
A tiered routing protocol minimizes LLM classification tokens by evaluating pattern matching, session state, and keyword lookups first. It only escalates ambiguous requests to the LLM when deterministic pre-checks fail, conserving computational resources.
Deterministic pre-checks are needed for request orchestration to guarantee minimal computation and safe fallback behavior. By resolving paths via pattern match and session state first, you avoid unnecessary LLM classification tokens and ensure predictable routing.
In this routing workflow, session state and keyword lookup act as intermediate decision tiers. The system checks active session data and matches specific keywords to resolve requests deterministically before falling back to LLM classification for remaining ambiguity.
The best way to handle ambiguous user requests is a tiered routing protocol with safe fallback behavior. It attempts pattern matching, session state, and keyword lookup first, escalating to LLM classification only when ambiguity remains.
Yes, you can use tiered routing for general routing scenarios. While it covers SDD-style workflows, the four-tier decision process applies to any orchestrator needing cost-optimized routing with deterministic pre-checks and safe fallback behavior.
salvo-rsConfigure Salvo session middleware with CookieStore or MemoryStore for user state.
VoTruongDanhAnalyze user requests to route tasks to specialist agents or orchestrators.
ryasrkAssign LLM model tiers to agents based on task complexity and reasoning needs.
troykellyTriage incoming work requests and route them to appropriate planning workflows.
