From hw
Configures global defaults for Hypo-Workflow agent platform, execution mode, subagents, plan mode, and dashboard via an interactive setup wizard.
How this skill is triggered — by the user, by Claude, or both
Slash command
/hw:setupThe summary Claude sees in its skill listing — used to decide when to auto-load this skill
📌 输出语言规则:
📌 输出语言规则: 读取 config.yaml → output.language
使用此技能作为插件级设置向导。它配置 Hypo-Workflow 本身,而不是项目本地的 .pipeline/ 工作区。
始终使用全局文件:
~/.hypo-workflow/~/.hypo-workflow/config.yaml不要从 setup 写入 .pipeline/config.yaml。项目本地配置属于 /hypo-workflow:init 或 /hypo-workflow:plan-generate。
当另一个技能需要运行时默认值时,按此顺序解析值:
.pipeline/config.yaml~/.hypo-workflow/config.yaml重要映射:
execution.mode > 全局 execution.default_mode > selfexecution.subagent_tool > 全局 subagent.provider > autoplan.mode > 全局 plan.default_mode > interactiveplan.interaction_depth > 全局 plan.interaction_depth > mediumdashboard.enabled > 全局 dashboard.enabled > falsedashboard.port > 全局 dashboard.port > 7700output.language > 全局 output.language > zh-CNoutput.timezone > 全局 output.timezone > Asia/Shanghaiwatchdog.enabled > 全局 watchdog.enabled > false.pipeline/rules.yaml > 项目 rules.* > 全局 rules.* > extends: recommendedmkdir -p ~/.hypo-workflow~/.hypo-workflow/config.yaml 不存在,启动首次运行向导claude-code 或 codexself 推荐给首次用户subagent 启用到另一个代理运行时的委托codex 或 claudebase_urltrue 或 false7700interactive 在规划期间暂停确认auto 运行规划阶段,无例行暂停zh-CNAsia/Shanghaifalse300recommended、strict 或 minimalrecommended~/.hypo-workflow/config.yaml。使用简短、具体的提示:
<detected>。使用此平台?"self 还是 subagent?"codex 还是 claude?"interactive 还是 auto?"low、medium 还是 high?"recommended、strict 还是 minimal?"agent.platform=claude-code 当在 Claude Code 中运行时,否则为 codexagent.model 当可见时应使用当前会话模型execution.default_mode=selfsubagent.provider=codexsubagent.codex.model=gpt-5.4subagent.claude.model=claude-sonnet-4-20250514dashboard.enabled=truedashboard.port=7700plan.default_mode=interactiveplan.interaction_depth=mediumplan.interactive.min_rounds=3plan.interactive.require_explicit_confirm=trueoutput.language=zh-CNoutput.timezone=Asia/Shanghaiwatchdog.enabled=falsewatchdog.interval=300watchdog.heartbeat_timeout=300watchdog.max_retries=5watchdog.max_consecutive_milestones=10watchdog.notify=truehistory_import.split_method=autohistory_import.time_gap_threshold=24hhistory_import.max_milestones=20compact.auto=truecompact.progress_recent=15compact.state_history_full=1compact.log_recent=20compact.reports_summary_lines=3showcase.language=autoshowcase.poster.api_key_env=OPENAI_API_KEYshowcase.poster.size=1024x1536showcase.poster.quality=highshowcase.poster.style=autorules.extends=recommendedrules.rules={}version=8.4.0# Hypo-Workflow global config
# Generated by /hypo-workflow:setup
agent:
platform: claude-code
model: claude-sonnet-4-20250514
execution:
default_mode: self
subagent:
provider: codex
codex:
model: gpt-5.4
base_url: https://api.vsplab.cn
claude:
model: claude-sonnet-4-20250514
dashboard:
enabled: true
port: 7700
plan:
default_mode: interactive
interaction_depth: medium
interactive:
min_rounds: 3
require_explicit_confirm: true
output:
language: zh-CN
timezone: Asia/Shanghai
watchdog:
enabled: false
interval: 300
heartbeat_timeout: 300
max_retries: 5
max_consecutive_milestones: 10
notify: true
history_import:
split_method: auto
time_gap_threshold: 24h
max_milestones: 20
keyword_patterns:
- 'feat\(M(\d+)\):'
- 'M(\d+)-'
- 'milestone-(\d+)'
compact:
auto: true
progress_recent: 15
state_history_full: 1
log_recent: 20
reports_summary_lines: 3
showcase:
language: auto
poster:
api_key_env: OPENAI_API_KEY
size: "1024x1536"
quality: high
style: auto
rules:
extends: recommended
rules: {}
version: "8.4.0"
created: "2026-04-26T14:00:00+08:00"
updated: "2026-04-26T14:00:00+08:00"
当使用默认 OpenAI 端点时,省略 subagent.codex.base_url。
仅运行与已安装工具匹配的检查:
codex --versionOPENAI_BASE_URL 和 OPENAI_API_KEYclaude --version如果检查失败,保留配置但在摘要中将提供者标记为未验证。除非用户要求严格检查,否则不要阻止设置。
Claude Code 用户:
hw 插件并调用命令为 /hw:*;Claude 原生 /resume 与 Hypo /hw:resume 分开@openai/codex、设置 OPENAI_API_KEY 并选择 subagent.provider=codex 将 Codex 配置为 subagentCodex 用户:
/hw:*.pipeline/config.yaml 中混合模式:
execution.mode=selfstep_overrides.<step>.executor=subagent 委托单个步骤当配置存在时:
updatedcreated 时间戳交互式配置编辑必须将全局默认值和项目配置作为单独的目标。首先暂存编辑,显示差异,验证支持的架构字段,并在写入前要求确认。
不要从 setup/config TUI 写入 .pipeline/state.yaml、.pipeline/cycle.yaml 或 .pipeline/rules.yaml。如果影响适配器的值发生变化,告诉用户运行 /hw:sync --light。
references/config-spec.md - 全局/项目配置优先级和字段映射config.schema.yaml - 项目架构加上全局配置架构定义SKILL.md - 完整系统参考(如果需要更广泛的上下文)npx claudepluginhub hypoxanthineovo/hypo-workflow --plugin hwValidates, runs, and debugs multi-agent YAML workflows. Use when orchestrating AI agents, configuring routing, or setting up human-in-the-loop gates.
Authors a reusable, deterministic agent orchestration workflow as a self-contained .mjs script for Claude Code's Workflow tool, with structured phases, fan-out, and verification.