一对匹配的Claude Code技能,根据会话主模型(Opus或Fable)动态调整行为,使用脚本检测当前模型并分配任务。

Stars

27

7 天增长

暂无数据

Fork 数

6

开放 Issue

0

开源协议

暂无数据

最近更新

2026-07-19

AI 仓库情报摘要
FR-AI / ANALYSIS

为什么值得关注

它引入了一种基于模型感知的智能架构,使得AI编码会话中不同模型之间可以无缝协调,支持自动重定向和升级模式。

适合谁使用

  • 使用多层级模型的Claude Code开发者
  • 构建自定义AI编码工作流的团队
  • 频繁在会话中切换模型的用户
  • 关注智能体编排模式的开发者

典型使用场景

  • 自动将复杂决策路由到更强大的模型(Opus),同时将大量工作委托给Sonnet
  • 将Fable作为前沿协调器,规划和委托任务给Codex和Sonnet工作器
  • 确保即使在会话中间切换模型后也能正确激活技能
  • 构建可扩展的多智能体编码助手,适应可用模型的强度

项目优势

  • 模型检测脚本能准确识别实时模型,即使会话中间切换且转录数据过时
  • 可共同安装的设计,自动重定向到正确的兄弟技能
  • 关注点清晰分离:advisor负责升级模式,coordinator负责前沿委托
  • 轻量级文件系统技能,无外部依赖

使用前须知

  • 仅当主模型为Opus或Fable时有效,对Sonnet或Haiku主导的会话无影响
  • 需要将两个技能文件夹同时安装到技能目录中

README 快速开始

fable-advisor + fable-coordinator

A matched pair of Claude Code skills encoding one subject — Fable 5's role in a session — split by the session's main model so each session loads only the half that applies.

SkillApplies when main model isRole
fable-advisorOpusOpus orchestrates, Sonnet does the volume workers, and a Fable subagent advises at genuine decision points (the escalation pattern — consult a stronger model on the hard calls).
fable-coordinatorFableFable is the frontier coordinator — plan big, execute small: keep only non-transferable in-session judgment, route self-contained frontier-grade legs to a Codex delegate, and fan mechanical legs out to Sonnet workers.

On a Sonnet- or Haiku-main session, neither applies.

Step-0 model probe

Both skills must know the session's current main model — and a model can't reliably report its own identity after a mid-session /model switch, because the conversation history keeps it anchored to what it was earlier. So each skill opens with a deterministic step 0 that derives the live model from the session transcript rather than trusting self-image, via the shared script fable-advisor/scripts/current-model.sh. It prints one token — opus, fable, sonnet, haiku, or unknown — and each skill branches on it, redirecting to its sibling if the wrong half loaded. The script is careful about transcript flush timing (a naive read of the last assistant message's model is stale exactly at step 0) and is hardened against tool-output that merely quotes model ids; the full rationale is in fable-advisor/README.md.

Install

Both are filesystem skills — drop the two folders into your skills directory (e.g. ~/.claude/skills/) so Claude Code discovers them at session start. The pair is designed to be co-installed: fable-coordinator calls the probe script that lives under fable-advisor/. See each folder's README.md for details and the automatic-firing setup.

项目描述

Claude skill: consult a stronger Fable 5 advisor subagent at decision points (Opus orchestrates, Sonnet workers, Fable advises).

相关仓库与替代方案

根据分类、Topic 和编程语言匹配的相似项目。

MoonshotAI
精选
MoonshotAI GitHub avatar

Kimi-K3

Kimi K3 is an open-weight, 2.8T-parameter native multimodal agentic model with a 1M-token context window, designed for frontier coding, knowledge work, and reasoning tasks.

AI 与机器学习AI 智能体
3,348
xuchonglang
精选
xuchonglang GitHub avatar

investing-for-beginners

A structured investing guide for Chinese beginners covering US stocks, options, and cryptocurrency, with focus on foundational concepts and risk awareness.

区块链与 Web3
2,739
Krishnagangwal
精选
Krishnagangwal GitHub avatar

CS-Fundamentals

A curated collection of Computer Science fundamentals (PDFs, notes, cheatsheets, interview question banks) for placement preparation, covering seven core subjects plus general resources.

数据与数据库数据库与存储
2,326