AgentSkillsCN

workflow

技能编排——区分用户指令与自动触发任务。

SKILL.md
--- frontmatter
name: workflow
description: Skill orchestration - user commands vs auto-triggered
user-invocable: false
model: haiku

Workflow Overview

User Commands (6)

code
┌─────────────────────────────────────────────────────────┐
│  "auto"          → Execute all tasks autonomously       │
│  "review"        → Code quality check on recent changes │
│  "brainstorm"    → Scan → propose → create stories      │
│  "test"          → npm test + browser tests on latest   │
│  "audit"         → Rate aspects → create stories        │
│  "status"        → Quick progress check                 │
└─────────────────────────────────────────────────────────┘

Aliases: "what next" = brainstorm

Auto-Triggered (Internal)

TriggerAction
Task completeverify runs automatically
Every 5 taskscheckpoint saves context
User says "ship"deploy runs
Artifacts pile upclean suggested

Analysis → Stories Flow

code
brainstorm / audit / what next
    ↓
Parallel scans (6 Haiku agents)
    ↓
Rate aspects / Present scenarios
    ↓
Auto-create stories for top issues
    ↓
User says "auto" → Execute

Task Lifecycle

code
TaskCreate (from brainstorm/audit)
    ↓
auto picks up task
    ↓
Implement → Typecheck → Build
    ↓
verify (auto) → Browser test if UX
    ↓
TaskUpdate: completed
    ↓
checkpoint (every 5) → Next task

Quick Reference

Want to...Say
Work through tasksauto
Check qualityreview
Don't know what's nextbrainstorm or what next
Run all teststest
Rate the appaudit
See progressstatus