Skip to main content

aim-model-dispatch

Select the appropriate LLM model for each agent based on task complexity and role

설치로 이동

소스 정보

저장소
Hidden-History/ai-memory
최근 소스 활동
2026년 8월 2일 01:05
감지된 SKILL.md 언어
영어
스타
41
포크
5

설치 방법

기본적으로 소스를 먼저 확인하는 Prompt가 선택됩니다. 직접 명령으로 전환하거나 로컬 사본을 다운로드할 수도 있습니다.

소스 파일 검토

설치 여부를 결정하기 전에 SKILL.md와 SkillsMP에 표시된 보조 파일을 읽어 보세요.

파일 탐색기
68 개 파일

SKILL.md 표시 중

SKILL.md
소스 지침 · 읽기 전용 미리보기
name
aim-model-dispatch
description
Select the appropriate LLM model for each agent based on task complexity and role
allowed-tools
Read
# Model Dispatch -- Model Selection and Provider Routing **Purpose**: Select the appropriate LLM model and route to the correct provider workflow. Called by /aim-agent-dispatch and /aim-agent-lifecycle. --- ## Model Selection ### By Complexity | Complexity | Model | |------------|-------| | Straightforward / Moderate | Sonnet | | Significant / Complex / Architectural | Opus | ### Role-Based Defaults | Agent Role | Default | Override When | |------------|---------|---------------| | DEV (implementation) | Sonnet | Opus if architectural changes or complex refactoring | | Code review | Sonnet | Opus if reviewing architectural decisions | | Analyst | Sonnet | Opus if deep architectural analysis | | PM | Sonnet | Opus if complex domain modeling | | Architect | Opus | Already at highest tier | | SM, UX Designer | Sonnet | Standard for all work | | Generic agent | Sonnet | Opus if task requires deep reasoning | ### Override Rules 1. **User override**: User preference always wins. 2. **Failed fix escalation**: After loop count > 1, consider upgrading to Opus. 3. **Haiku**: Only for simple, high-volume parallel tasks (file scanning, grep-and-report). Never for implementation, review, or planning. 4. **Cost awareness**: Use Opus when reasoning depth justifies the cost, not as default. 5. **Non-Claude providers**: Reasoning tier (Opus/Sonnet/Haiku) maps to equivalent model on the selected provider. --- ## Provider Routing ### Path 1: Claude Native **MANDATORY** for opus, sonnet, haiku -- use the Claude Code native parallel system. → [`workflows/claude-native/workflow.md`]({skills_path}/aim-model-dispatch/workflows/claude-native/workflow.md) ### Path 2: Non-Claude Providers | Provider | Workflow | |----------|----------| | openrouter, ollama, gemini, deepseek, groq, cerebras, mistral, openai, vertex-ai, siliconflow | [`tmux-dispatch`]({skills_path}/aim-model-dispatch/workflows/tmux-dispatch/workflow.md) or [`bmad-dispatch`]({skills_path}/aim-model-dispatch/workflows/bmad-dispatch/workflow.md) | | api (image/audio/video) | [`api-dispatch`]({skills_path}/aim-model-dispatch/workflows/api-dispatch/workflow.md) | --- ## Framework All sub-workflows, references, scripts, wrappers, and evals are in: `{skills_path}/aim-model-dispatch/`
GitHub에서 보기