AgentSkillsCN

skill-flow-orchestrator

一种元技能，可将其他技能编排成智能且自我协调的工作流。当复杂任务需要多个技能协同工作、需根据任务分析实现自动化技能路由、希望保存并重放成功的技能组合、需要在不同技能边界间保持状态持久化，或构建多阶段处理流水线时，可使用此功能。提供四种模式：快速模式（1–2 分钟，单个技能）、标准模式（5–10 分钟，多技能链）、深度模式（15–30 分钟，DAG 组合）、专家模式（自定义模式，支持学习与优化）。

SKILL.md
--- frontmatter
name: skill-flow-orchestrator
description: Meta-skill that orchestrates other skills into intelligent, self-coordinating workflows. Use when complex tasks require multiple skills working together, need automated skill routing based on task analysis, want to save and replay successful skill combinations, need state persistence across skill boundaries, or building multi-stage processing pipelines. Provides 4 modes - QUICK (1-2min, single skill), STANDARD (5-10min, multi-skill chains), DEEP (15-30min, DAG composition), EXPERT (custom, learning optimization).
version: 1.0.0

Skill-Flow-Orchestrator (SFO)

Production-ready Skill Orchestration System that transforms isolated skills into an intelligent, self-coordinating ecosystem.

Core Capabilities

  1. •Intelligent Routing - Match task DNA to skill DNA for optimal selection
  2. •Flow Composition - Build skill chains as Directed Acyclic Graphs (DAGs)
  3. •Context Management - Persist state across skill executions (hybrid: memory file + in-prompt)
  4. •Chain Templates - Save successful chains as reusable workflows
  5. •Adaptive Learning - Optimize routing based on execution outcomes

When to Use

  • •Complex tasks: Require multiple skills working together
  • •Automated selection: Need intelligent skill routing based on task analysis
  • •Workflow reuse: Save and replay successful skill combinations
  • •Context continuity: Need state persistence across skill boundaries
  • •Pipeline automation: Building multi-stage processing workflows

4 Orchestration Modes

ModeDurationFeaturesBest For
QUICK1-2 minBasic routing, single skillSimple task delegation
STANDARD5-10 minMulti-skill chains, context passingCommon workflows
DEEP15-30 minFull DAG composition, parallel executionComplex pipelines
EXPERTCustomLearning optimization, template managementProduction workflows

Quick Start

1. Scan Available Skills

bash
python3 .agent/skills/skill-flow-orchestrator/scripts/scan_skills.py

Output: .agent/sfo/skill_index.json - indexed skills with DNA

2. Match Skills to Task

bash
python3 .agent/skills/skill-flow-orchestrator/scripts/match_skills.py "<task_description>"

Returns ranked skills with match scores

3. Orchestrate Workflow

bash
python3 .agent/skills/skill-flow-orchestrator/scripts/orchestrate.py --mode standard "<task>"

Orchestration Workflow

  1. •INTAKE - Parse task, extract requirements
  2. •ANALYZE - Match task DNA with skill DNA using semantic matching
  3. •COMPOSE - Build execution DAG from matched skills
  4. •EXECUTE - Run skills in order, passing context
  5. •ADAPT - Log outcomes, update routing weights

Context Management

  • •Long-term state: .agent/sfo/context.json
  • •Chain templates: .agent/sfo/templates/
  • •Execution logs: .agent/sfo/logs/

References