spawn-task

name	spawn-task
description	sub-agent fork 模板：给 prompt 模板 + 输出格式约定 + 资源约束 + 完成检测，fork 一个 Agent sub-task 后台跑。比每次手写 Agent prompt 省 10x。Use when user says "spawn task", "fork agent", "开个 sub-agent", "并行任务", "background agent", "派活".
argument-hint	[task-description] [--type research|review|code|audit] [--background]
allowed-tools	Agent, Read, Write

/spawn-task — Sub-Agent Fork Template

借鉴 ARIS /experiment-queue + AutoResearchClaw sub-agent pattern。

Overview

把「fork 一个 sub-agent 让它去搞 X」标准化：

字段	内容
task_description	一句话
output_format	JSON schema 或 markdown 段落约定
resource_constraint	"max 10 tool calls" 或 "no WebSearch"
completion_signal	sub-agent 写 .auto-production/spawn/<id>.done
timeout	默认 30min

Workflow

ID=$(uuidgen | head -c 8)
mkdir -p .auto-production/spawn/$ID

# emit prompt template
python3 tools/spawn_task.py prompt --type $TYPE --task "$TASK" > .auto-production/spawn/$ID/prompt.md

# fork via Agent tool (foreground or --background)
Agent(prompt=prompt_text, description="$TASK", run_in_background=$BG)

# completion: sub-agent should `touch .done`

5 个 built-in --type：

• research — survey + summary
• review — code review
• code — implement a single function
• audit — fact-check / cite-verify
• freeform — user 自定 prompt

Output

• .auto-production/spawn/<id>/prompt.md
• .auto-production/spawn/<id>/result.md
• .auto-production/spawn/<id>/.done

Composition

• /research-pipeline 调多次 spawn (并行 idea-perspective + persona-probe)
• /audit-driven-retrain spawn audit sub-agent
• /cross-review spawn 2 critic sub-agent