npx skills add https://github.com/willsigmon/sigstack --skill 'Spawn Swarm'并行代理执行并行工作。
Single message → Multiple Task calls → Parallel execution → Collected results
一条消息包含 5-20 个任务工具调用。所有任务同时运行。结果一并返回。
"在整个代码库中查找缺陷"
按模块生成:
- auth-agent
- payment-agent
- user-agent
- api-agent
- data-agent
"审查 PR #123"
生成专家:
- security-reviewer
- performance-reviewer
- style-reviewer
- test-coverage-reviewer
"构建 [功能]"
生成工作器:
- ui-agent
- logic-agent
- test-agent
- docs-agent
"理解 [某事物] 的工作原理"
生成探索者:
- entry-point-finder
- flow-tracer
- dependency-mapper
- pattern-identifier
- edge-case-hunter
广告位招租
在这里展示您的产品或服务
触达数万 AI 开发者,精准高效
[Task: Search auth/ for security issues]
[Task: Search payment/ for security issues]
[Task: Search api/ for security issues]
[Task: Search data/ for security issues]
[Task: Search user/ for security issues]
全部在一条消息中。
对于长时间运行的工作:
run_in_background: true
继续其他工作。
稍后使用 TaskOutput 检查结果。
快速文件/代码搜索
模式查找
代码库问题解答
多步骤实现
复杂变更
自主工作
code-reviewer
debugger
security-reviewer
performance-reviewer
| 任务 | 代理数量 | 原因 |
|---|---|---|
| 小型搜索 | 3-5 | 快速覆盖 |
| 完整代码库 | 5-10 | 每个主要区域一个 |
| 深度分析 | 10-15 | 彻底探索 |
| 大规模重构 | 15-20 | 最大并行度 |
主对话:完整代码库
每个代理:仅其目标区域
5 个代理搜索 5 个区域:
- 令牌数与 1 个代理完成全部 5 项相同
- 速度快 5 倍
- 每个代理都有专注的上下文
代理进行广泛探索。
返回:"在 auth.swift:142 处发现问题"
您获得一行摘要。
而非完整的探索跟踪记录。
- 为简单读取生成代理(直接使用 Read 工具)
- 生成有依赖关系的代理(竞态条件)
- 生成 50 个代理(混乱)
- 在生成下一个之前等待每个代理
- 一条消息,所有代理
- 独立的目标
- 最多 5-20 个代理
- 长时间工作使用后台运行
生成一个缺陷搜寻集群:
1. 搜索 src/auth/ 查找安全漏洞
2. 搜索 src/payment/ 查找数据处理问题
3. 搜索 src/api/ 查找输入验证缺陷
4. 搜索 src/user/ 查找权限问题
5. 搜索 src/data/ 查找竞态条件
按严重性等级返回发现的问题。
使用时机:任何并行工作的机会
每周安装次数
–
代码仓库
GitHub 星标数
8
首次出现时间
–
安全审计
Parallel agents for parallel work.
Single message → Multiple Task calls → Parallel execution → Collected results
One message with 5-20 Task tool invocations. All run simultaneously. Results come back together.
"Find bugs across the codebase"
Spawn per module:
- auth-agent
- payment-agent
- user-agent
- api-agent
- data-agent
"Review PR #123"
Spawn specialists:
- security-reviewer
- performance-reviewer
- style-reviewer
- test-coverage-reviewer
"Build [feature]"
Spawn workers:
- ui-agent
- logic-agent
- test-agent
- docs-agent
"Understand how [thing] works"
Spawn explorers:
- entry-point-finder
- flow-tracer
- dependency-mapper
- pattern-identifier
- edge-case-hunter
[Task: Search auth/ for security issues]
[Task: Search payment/ for security issues]
[Task: Search api/ for security issues]
[Task: Search data/ for security issues]
[Task: Search user/ for security issues]
All in ONE message.
For long-running work:
run_in_background: true
Continue other work.
Check results with TaskOutput later.
Fast file/code search
Pattern finding
Codebase questions
Multi-step implementation
Complex changes
Autonomous work
code-reviewer
debugger
security-reviewer
performance-reviewer
| Task | Agents | Why |
|---|---|---|
| Small search | 3-5 | Quick coverage |
| Full codebase | 5-10 | One per major area |
| Deep analysis | 10-15 | Thorough exploration |
| Massive refactor | 15-20 | Maximum parallelism |
Main conversation: Full codebase
Each agent: Only its target area
5 agents searching 5 areas:
- Same tokens as 1 agent doing all 5
- 5x faster
- Each has focused context
Agent explores extensively.
Returns: "Found issue in auth.swift:142"
You get 1-line summary.
Not the full exploration trace.
- Spawn for simple reads (just use Read tool)
- Spawn with dependencies (race conditions)
- Spawn 50 agents (chaos)
- Wait on each before spawning next
- One message, all agents
- Independent targets
- 5-20 agents max
- Background for long work
Spawn a bug-hunting swarm:
1. Search src/auth/ for security vulnerabilities
2. Search src/payment/ for data handling issues
3. Search src/api/ for input validation bugs
4. Search src/user/ for permission issues
5. Search src/data/ for race conditions
Return findings ranked by severity.
Use when: Any parallel work opportunity
Weekly Installs
–
Repository
GitHub Stars
8
First Seen
–
Security Audits
AI Elements:基于shadcn/ui的AI原生应用组件库,快速构建对话界面
58,500 周安装