agent-contracts-routing-tuning

스타12

포크0

업데이트2026년 1월 17일 10:31

Tune routing with TriggerCondition priorities, LLM hints, context_builder, and traceable decisions.

설치

Codex 또는 Claude로 설치 이 Prompt를 복사해 Codex, Claude 또는 다른 어시스턴트에 붙여 넣으면 Skill 페이지를 검토하고 설치를 진행할 수 있습니다.

출처

yatarousan0227

yatarousan0227/agent-contracts

GitHub 저장소 열기 Creator 저장소 보기

다운로드

agent-contracts Routing Tuning

Use this skill when you are designing or debugging routing behavior (rule matches vs LLM selection).

Routing Model (what to optimize)

Rule filtering via TriggerCondition (when / when_not, priority)
Candidate selection (top matches + ties)
Optional LLM decision among candidates (with llm_hint)
Fallbacks and terminal states (response.response_type)

Practical Tuning Steps

Make rule-based selection deterministic first (LLM off).
Use priorities to express business rules (100+ = critical, 50-99 = main, 1-49 = fallback).
Add llm_hint only where ambiguity remains after rules.
Use context_builder only when needed; keep default minimal slices.
Debug with decide_with_trace() and inspect matched rules.

Guardrails

Prefer rules for safety/constraints; use LLM for ambiguous intent.
Keep candidate sets small and explainable.
Treat response.response_type terminal values as part of routing design.

References (load only when needed)

docs/core_concepts.md (Traceable Routing / Context Builder)
examples/02_routing_explain.py
docs/skills/official/agent-contracts-routing-tuning/references/debug_playbook.md

이 저장소의 다른 Skills

같은 저장소

agent-contracts-app-builder

yatarousan0227/agent-contracts

Build a production-minded AI agent using agent-contracts (contracts, slices, validation, graph, runtime, CLI tooling).

2026-01-1712

agent-contracts-backend-runtime

yatarousan0227/agent-contracts

Build an API-oriented agent using AgentRuntime/StreamingRuntime with predictable request/response slices and session support.

2026-01-1712

agent-contracts-contracts-ci

yatarousan0227/agent-contracts

Make contract-driven agents safe to change with strict validation, architecture docs, and contract diffs in CI.

2026-01-1712

agent-contracts-interactive-flow

yatarousan0227/agent-contracts

Build question/answer style agents using InteractiveNode, explicit routing, and workflow slices.

2026-01-1712

agent-contracts-tool-use-rag

yatarousan0227/agent-contracts

Implement tool-using or retrieval-style nodes with services, dependency injection, and safe LLM context building.

2026-01-1712

name	agent-contracts-routing-tuning
description	Tune routing with TriggerCondition priorities, LLM hints, context_builder, and traceable decisions.
metadata	{"short-description":"Routing design and debugging"}

agent-contracts Routing Tuning

Use this skill when you are designing or debugging routing behavior (rule matches vs LLM selection).

Routing Model (what to optimize)

Rule filtering via TriggerCondition (when / when_not, priority)
Candidate selection (top matches + ties)
Optional LLM decision among candidates (with llm_hint)
Fallbacks and terminal states (response.response_type)

Practical Tuning Steps

Make rule-based selection deterministic first (LLM off).
Use priorities to express business rules (100+ = critical, 50-99 = main, 1-49 = fallback).
Add llm_hint only where ambiguity remains after rules.
Use context_builder only when needed; keep default minimal slices.
Debug with decide_with_trace() and inspect matched rules.

Guardrails

Prefer rules for safety/constraints; use LLM for ambiguous intent.
Keep candidate sets small and explainable.
Treat response.response_type terminal values as part of routing design.

References (load only when needed)

docs/core_concepts.md (Traceable Routing / Context Builder)
examples/02_routing_explain.py
docs/skills/official/agent-contracts-routing-tuning/references/debug_playbook.md