openclaw 网盘下载
OpenClaw

技能详情(站内镜像,无评论)

首页 > 技能库 > Parallel Agents

Use when facing 2 or more independent tasks that can be worked on without shared state - dispatches parallel subagents using sessions_spawn for concurrent in...

开发与 DevOps

许可证:MIT-0

MIT-0 ·免费使用、修改和重新分发。无需归因。

版本:v1.0.1

统计:⭐ 0 · 42 · 0 current installs · 0 all-time installs

0

安装量(当前) 0

🛡 VirusTotal :良性 · OpenClaw :良性

Package:axelhu/superpowers-parallel-agents

安全扫描(ClawHub)

  • VirusTotal :良性
  • OpenClaw :良性

OpenClaw 评估

The skill is internally consistent: it provides only instructions to spawn parallel subagents for independent tasks, requests no installs or credentials, and its behavior aligns with its description.

目的

The name/description (dispatch parallel subagents for independent tasks) matches the SKILL.md content. The instructions only require the agent to create sessions with sessions_spawn, read test files, run tests, and propose or apply fixes — all coherent with the stated purpose.

说明范围

The instructions explicitly direct subagents to read project files, run tests, and make code fixes (with constraints such as 'don't change other code'). This is expected for a code-fixing parallel-agents skill, but it does mean the skill involves filesystem access and code modification. The guidance is high-level and relies on the orchestrating agent to enforce constraints and resolve merge/conflict risks, so users should ensure human review/c…

安装机制

Instruction-only skill with no install spec and no code files. Nothing is downloaded or written to disk by an installer, which is the lowest-risk install surface.

证书

No required environment variables, credentials, or config paths are declared, and the SKILL.md does not request secrets or external tokens. The described operations (sessions_spawn, local cwd) do not require additional credentials beyond whatever the agent already has to access the project workspace.

持久

always is false and there is no attempt to modify other skills or system-wide settings. Autonomous invocation remains allowed (platform default) but the skill itself does not demand elevated or persistent privileges.

综合结论

This skill appears coherent and limited to orchestrating parallel subagents, but it will enable agents to read and modify your codebase — so take precautions: run it on a cloned repo or feature branch, require a human review step before merging any automated changes, ensure CI runs the full test suite after integration, and watch for conflicting edits from concurrent agents. Do not run it against repositories containing sensitive secrets or pr…

安装(复制给龙虾 AI)

将下方整段复制到龙虾中文库对话中,由龙虾按 SKILL.md 完成安装。

请把本段交给龙虾中文库(龙虾 AI)执行:为本机安装 OpenClaw 技能「Parallel Agents」。简介:Use when facing 2 or more independent tasks that can be worked on without share…。
请 fetch 以下地址读取 SKILL.md 并按文档完成安装:https://raw.githubusercontent.com/openclaw/skills/refs/heads/main/skills/axelhu/superpowers-parallel-agents/SKILL.md
(来源:yingzhi8.cn 技能库)

SKILL.md

打开原始 SKILL.md(GitHub raw)

---
name: superpowers-parallel-agents
description: Use when facing 2 or more independent tasks that can be worked on without shared state - dispatches parallel subagents using sessions_spawn for concurrent investigation and execution, adapted for OpenClaw
---

# Superpowers Parallel Agents(OpenClaw 适配版)

## 概述

当面临 2+ 个独立任务时,将它们分配给专门的并行 subagent,同时工作。OpenClaw 用 `sessions_spawn` 创建独立 session 实现并行分发。

**核心原则:** 每个独立问题域分配一个 agent,让它们并发工作。

## OpenClaw 适配

Superpowers 原版用 Claude Code 的 `Task` 工具并发派发。OpenClaw 用 `sessions_spawn`:
- `sessions_spawn(mode="run")` — 一次性任务,并发执行
- `sessions_spawn(mode="session")` — 持久 session,可多轮交互
- 主 session 协调,subagent 结果通过 session 历史或文件系统汇总

## 使用条件

```
有多个失败/独立任务?
  → 它们独立?(不同根因、无共享状态)?
    → 可以并行工作?
      → 用 parallel-agents(这个技能)✅
    → 顺序更合适?
      → 用 systematic-debugging 单独处理
  → 需要全面上下文理解?
    → 单一 agent 处理
```

**适用场景:**
- 3+ 个测试文件失败,根因不同
- 多个子系统独立损坏
- 每个问题可以在不理解其他问题上下文的情况下理解
- 调查之间无共享状态

**不适用:**
- 失败互相关联(修一个可能修其他)
- 需要理解完整系统状态
- Agents 会互相干扰(编辑同一文件、共用资源)

## 流程

### 1. 识别独立问题域

按问题分组:
- 测试文件 A:工具审批流程
- 测试文件 B:批处理完成行为
- 测试文件 C:中止功能

每个问题域独立——修工具审批不影响中止测试。

### 2. 为每个 Agent 创建专注任务

每个 agent 获得:
- **明确范围:** 一个测试文件或子系统
- **清晰目标:** 让这些测试通过 / 修复这个 bug
- **约束:** 不要改其他代码
- **预期输出:** 发现什么、修复什么的摘要

### 3. 并行 dispatch

用 `sessions_spawn` 同时派发所有 agent:

```javascript
// OpenClaw: sessions_spawn 并行派发
sessions_spawn({
  task: "修复 src/agents/agent-tool-abort.test.ts 的 3 个失败测试...",
  runtime: "subagent",
  mode: "run",
  cwd: "/path/to/project"
})
sessions_spawn({
  task: "修复 src/batch/completion.test.ts 的 2 个失败测试...",
  runtime: "subagent",
  mode: "run",
  cwd: "/path/to/project"
})
sessions_spawn({
  task: "修复 src/tools/race-conditions.test.ts 的 1 个失败测试...",
  runtime: "subagent",
  mode: "run",
  cwd: "/path/to/project"
})
```

### 4. 审查和整合

当 agents 返回:
- 读每个摘要
- 验证修复不冲突
- 运行完整测试套件
- 整合所有变更

## Agent 提示词结构

好的 agent 提示词:
1. **专注** — 一个清晰的问题域
2. **自包含** — 理解问题所需的全部上下文
3. **输出具体** — agent 应该返回什么?

```markdown
修复 src/agents/agent-tool-abort.test.ts 中 3 个失败的测试:

1. "should abort tool with partial output capture" - 期望消息中有 'interrupted at'
2. "should handle mixed completed and aborted tools" - 快速工具被中止而非完成
3. "should properly track pendingToolCount" - 期望 3 个结果但得到 0

这些是时序/竞态条件问题。你的任务:

1. 读测试文件,理解每个测试验证什么
2. 识别根因——时序问题还是实际 bug?
3. 修复:
   - 用事件等待替代任意 timeout
   - 如发现 bug 则修复 abort 实现
   - 如测试的是变化的行为则调整测试期望

不要只加 timeout——找真正的问题。

返回:你发现了什么,修复了什么。
```

## 常见错误

**❌ 范围太广:** "修所有测试" — agent 会迷失
**✅ 具体:** "修 agent-tool-abort.test.ts" — 专注范围

**❌ 无上下文:** "修竞态条件" — agent 不知道在哪里
**✅ 有上下文:** 粘贴错误信息和测试名

**❌ 无约束:** Agent 可能重构一切
**✅ 有约束:** "不要改其他代码" 或 "只修测试"

**❌ 输出模糊:** "修好了" — 不知道改了啥
**✅ 输出具体:** "返回根因和变更摘要"

## 何时不用

**相关失败:** 修一个可能修其他——先一起调查
**需要完整上下文:** 理解需要看到整个系统
**探索性调试:** 还不知道哪里坏了
**共享状态:** Agents 会互相干扰(编辑同一文件、用同一资源)

## 关键优势

1. **并行化** — 多个调查同时进行
2. **专注** — 每个 agent 范围窄,跟踪的上下文少
3. **独立性** — Agents 不互相干扰
4. **速度** — 3 个问题用 1 个问题的时间解决

## 验证

Agents 返回后:
1. **审查每个摘要** — 理解改了什么
2. **检查冲突** — Agents 编辑了同一代码吗?
3. **运行完整套件** — 验证所有修复一起工作
4. **抽查** — Agents 可能犯系统性错误