Token导航 LogoToken导航

Agent Skill · 研究检索

popcorn-xp(爆米花 xp)

popcorn-xp 用于查找、检索和筛选相关信息,适合在 Codex、Claude、Cursor、Gemini CLI 中需要根据关键词、任务场景或来源线索快速定位候选结果时使用。可结合来源仓库、安装命令和原始 README 继续核验具体用法。安装前建议确认权限范围、维护状态,以及是否会触发联网、命令执行或文件读写。

popcorn-xp 是研究检索类 Agent Skill,可用宿主为 Codex、Claude、Cursor、Gemini CLI,通过GitHub 仓库获取,权限为执行命令;许可证信息还不明确,第三方扫描有提醒项,按安装量在研究检索类的 73,675 个 Skill 里排第 66,030。

总安装 269周安装 11GitHub Stars 公开资料未说明下载 86核验 2026-05-01

安装说明

本站只整理中文说明,不托管安装包,也不代用户安装。
GitHub
许可证
unknown

安装方式

通过对话安装

复制提示词发给支持本地命令或 Skills 的 AI 助手,先确认命令和权限,再让它执行。

请帮我安装这个 Agent Skill:popcorn-xp(爆米花 xp)
来源仓库:https://github.com/mikewolfd/xp-popcorn-skill
仓库路径:skills/popcorn-xp
安装命令:
npx skills add https://github.com/mikewolfd/xp-popcorn-skill --skill popcorn-xp
安装前请先检查当前环境是否支持对应 CLI,并向我确认将要执行的命令、安装目录、联网范围和文件读写权限;确认后再执行。

命令行安装

复制命令到本机终端执行。该命令会通过 npx skills 从第三方来源获取 Skill;本站只展示命令,不托管安装包,也不自动执行。

skills.shnpx skills
npx skills add https://github.com/mikewolfd/xp-popcorn-skill --skill popcorn-xp

这个 Skill 是做什么的

研究检索类 Skill 主要解决「找资料、核对事实、把来源整理清楚」的问题,通常会让助手先检索再总结。

爆米花 xp 用于查找、检索和筛选相关信息,适合在 Codex、Claude、Cursor、Gemini CLI 中快速定位候选结果。它支持根据关键词或任务场景进行信息筛选,适用于经验值相关场景的数据调研。

可通过 npx skills add 命令从指定 GitHub 仓库安装,具体用法需结合原始 README 进一步确认。安装前建议核实权限范围、维护状态,并评估是否会触发联网或文件操作。popcorn-xp 属于研究检索类 Skill,可作为该场景下的辅助能力补充。

来源仓库内路径:skills/popcorn-xp

上手步骤

4 步
  1. 1

    确认自己的宿主工具

    公开资料标注支持 Codex、Claude、Cursor、Gemini CLI;先确认自己用的是其中之一,再按对应工具的 Skills 目录放置。

  2. 2

    拿到安装命令

    可以直接复制的命令是 npx skills add https://github.com/mikewolfd/xp-popcorn-skill --skill popcorn-xp;命令里的占位符换成自己的路径或密钥再执行。

  3. 3

    安装后重启宿主

    命令执行完要完全退出宿主工具再打开,让新装的 Skill 被重新扫描到,否则列表里可能看不到。

  4. 4

    按最小权限试用

    它需要执行命令,先在测试目录或测试账号里跑,确认没问题再放到正式项目。

常见问题

6 条

popcorn-xp 怎么安装?

在终端执行 npx skills add https://github.com/mikewolfd/xp-popcorn-skill --skill popcorn-xp。装完完全退出宿主工具再打开,让它重新扫描本地 Skill。

popcorn-xp 支持哪些 AI 工具?

公开资料标注支持 Codex、Claude、Cursor、Gemini CLI。其他工具只要能读取本地 Skill 目录,一般也能手动放进去,但要自己确认触发方式是否一致。

popcorn-xp 需要什么权限?

权限级别是执行命令。安装流程涉及命令执行,可能通过 npx skills add https://github.com/mikewolfd/xp-popcorn-skill --skill popcorn-xp 联网下载 Skill 或依赖。用户安装前应确认命令来源、仓库内容和执行环境。

popcorn-xp 是免费的吗?

公开资料没有标注明确许可证,收费和商用限制都无法确认。正式使用前请到来源仓库核对授权说明。

popcorn-xp 用的人多吗?

公开数据是累计安装 269 次,近一周安装 11 次,下载 86 次。这些数字来自第三方目录统计,只能当作热度参考,不代表质量或安全性。

popcorn-xp 安全吗?

第三方扫描有提醒项,说明存在需要人工确认的行为。安装前请打开源码看清它读了什么、联网到哪里。

值得选的地方和要先想清楚的地方

值得选的地方

  • 同时适配 4 种工具,换客户端不用重新找替代品。

先想清楚的地方

  • 需要执行命令,装到常用环境前要先评估影响面。
  • 许可证没有明确标注,能不能商用、能不能再分发都不确定。
  • 第三方扫描存在提醒项或未通过项,需要人工逐条确认。
  • 第三方目录整理,版本更新和行为细节以来源仓库为准。

适不适合你

更适合

  • 已经在用 Codex,想直接扩展它能力的人。
  • 需要处理研究检索类任务、又不想每次都手写提示词的人。

不太适合

  • 不能接受 Skill 在你机器上执行命令的人。
  • 需要明确授权才能用到公司项目里的场景。
  • 需要厂商级技术支持和 SLA 的生产系统。

研究检索类 Skill 对比

6 个同类
Skill总安装GitHub 星权限许可证扫描
find-skills

寻找技能 · Codex、Claude

31,199,81316,486操作浏览器MIT有提醒
self-improving-agent

自我改进 Agent · OpenClaw

16,464,0023,396接触敏感数据MIT-0通过
soultrace

灵魂追踪 · Codex、Claude

7,588,94412需要联网未明确有提醒
self-improving

自我完善 · OpenClaw

6,829,2841,046需要联网MIT-0通过
microsoft-foundry

微软代工厂 · Codex、Claude

6,614,144764调用外部服务未明确有提醒
popcorn-xp

爆米花 xp · 本页

269—执行命令未明确有提醒

按安装量取同分类前几名;数字来自第三方目录统计,只作热度参考,不代表质量结论。

SKILL.md 原文

Popcorn XP

Launch an XP pair-programming session. You (the lead) set up the team and step back. Teammates pair-program directly with each other via SendMessage. One driver edits, one navigator steers, they swap roles between tasks. Advice has teeth — OBJECTIONs block task completion.

Trigger

Activate when the user explicitly asks for a team-style workflow:

  • "pair program on this"
  • "run an XP session"
  • "use subagents"
  • "let a team of agents work this"
  • "popcorn this task"

Do not activate for ordinary single-agent coding.

Role Roster

The plugin ships with agent definitions in agents/ that can be used as teammates. Pick 2-3 agents that match the task. Default to the core four for coding tasks; add specialists when the task calls for them.

Core roles (coding tasks):

AgentLensModel
popcorn-xp:scout"Are we solving the right problem?"sonnet
popcorn-xp:craftsman"Is this clean and readable?"opus
popcorn-xp:expert"Does this actually work in edge cases?"opus (has project memory)
popcorn-xp:tester"How will we prove this?"opus

Specialist roles (when needed):

AgentLensModel
popcorn-xp:service-designer"Does the interface serve the experience — API to UI?"sonnet
popcorn-xp:visual-designer"Does this look right and feel right?"sonnet
popcorn-xp:qa"Does this work from the user's perspective?"sonnet
popcorn-xp:product-manager"What problem are we solving, and is this the right way?"sonnet

Checking for local overrides: Before spawning teammates, check if the project has custom agents in .claude/agents/ that overlap with the popcorn-xp roles (e.g., a project-specific test-engineer or code-scout). Prefer local agents when they exist — they carry project-specific context. Use popcorn-xp agents as defaults when no local override exists.

The lens shapes how an agent thinks, not what it's allowed to do. Any teammate can drive, navigate, write tests, or review code. When rotating, prefer giving the driver role to whoever was just navigating — they carry context from watching the code emerge.

Workflow

1. Understand the Task

Before creating the team, understand the problem yourself. If you have file access, read the relevant code. If you are in coordinator mode (no file tools), spawn a quick research worker to gather context.

Check for prior retros. If .popcorn-xp/RETRO.md exists, read it. It contains process observations from previous sessions on this codebase — what worked, what didn't, what to change. Apply any relevant recommendations.

Build a mental model of:

  • What files are involved
  • What the user wants changed
  • What could go wrong
  • How to verify success

Break the work into 3-6 concrete tasks.

2. Create the Team

TeamCreate "popcorn-xp"

3. Create Tasks

Use TaskCreate for each work item. Set dependencies with TaskUpdate so tasks unblock in order.

Example task breakdown:

Task 1: "Map affected files, entry points, and constraints"
Task 2: "Implement depth validation in parseBlock()" — blocked by 1
Task 3: "Add regression tests for invalid input" — blocked by 2
Task 4: "Run full test suite and verify no regressions" — blocked by 3

Include enough context in each task description for a teammate to execute it independently. State what to do, why it matters, and what success looks like.

4. Spawn Teammates

Launch 2-3 teammates using the Agent tool with team_name: "popcorn-xp". Give each teammate the protocol from references/protocol.md and their role assignment.

Driver teammate — the role best suited to the first task:

Agent(
  name: "craftsman",
  team_name: "popcorn-xp",
  prompt: "<driver coordinator prompt from protocol.md>"
)

Navigator teammate — the role that provides the most useful counter-lens:

Agent(
  name: "expert",
  team_name: "popcorn-xp",
  prompt: "<navigator coordinator prompt from protocol.md>"
)

Advisor teammate (optional) — when a third perspective adds value:

Agent(
  name: "tester",
  team_name: "popcorn-xp",
  prompt: "<advisor coordinator prompt from protocol.md>"
)

Assign the first task to the driver via TaskUpdate.

5. Monitor

You receive messages from teammates automatically. Your role during execution:

  • Acknowledge completion messages. When a teammate finishes a task, check TaskList and assign or approve the next task.
  • Steer when needed. If a teammate is going in the wrong direction, SendMessage with guidance.
  • Relay user input. If the user provides new instructions, SendMessage to the relevant teammate.
  • Handle rotation. When a task completes, swap the driver and navigator roles. The agent that was navigating should drive the next task — they've been watching the code and carry context the other agent doesn't. Resist assigning tasks to the "best-fit" role; rotation is for knowledge sharing, not specialization.
  • Handle escalations. If the navigator sends an ESCALATION message (the approach is fundamentally wrong), pause the current task, evaluate the concern, and decide whether to redirect, reset, or continue.
  • Do not do the work yourself. You are the lead, not a driver. If you find yourself wanting to read a file or write code, delegate it to a teammate instead.

6. Verify and Close

When all tasks are complete:

  1. Ask a teammate (typically the tester) to run final verification.
  2. Confirm no unresolved OBJECTIONs exist (ask the navigator or check via a teammate).
  3. Retrospective — ask each teammate: "What worked well? What would you change about the process? Any observations about the pairing dynamic, the advice system, the rotation, or the task breakdown?" Collect their responses.
  4. Present a technical summary to the user: what was done, what each role found, any remaining risk.
  5. Shut down teammates: SendMessage(to: "*", message: {type: "shutdown_request"})
  6. After teammates shut down: TeamDelete
  7. Write the retro file. After TeamDelete, write .popcorn-xp/RETRO.md with your assessment of the session. This is YOUR perspective as the lead — what you observed about how the team worked, not just what they built. Use the format below.

Retro File Format

Write .popcorn-xp/RETRO.md after every session. This file accumulates across sessions — append a new entry, don't overwrite prior retros.

# Popcorn XP Retro

## Session: {date} — {task summary}

### Team
- Driver(s): {who drove which tasks}
- Navigator(s): {who navigated which tasks}
- Advisor(s): {if any}

### What Worked
- {concrete observation — e.g., "rotation after task 2 gave the expert context they used to catch OBJ-3-01"}
- {concrete observation}

### What Didn't Work
- {concrete observation — e.g., "navigator went idle for 3 checkpoints because checkpoints were too small to analyze"}
- {concrete observation}

### Advice System
- OBJECTIONs raised: {count}
- OBJECTIONs fixed: {count}
- OBJECTIONs rejected: {count}
- SMELLs/STEERs/FYIs: {count}
- Assessment: {did the advice system create the right dynamic? too many objections? too few? did the driver push back enough?}

### Rotation
- {did rotation spread knowledge? did assigning by lens-fit happen despite the protocol? did the navigator-becomes-driver pattern work?}

### Process Observations
- {anything about the protocol itself — too much ceremony? not enough checkpoints? file format issues?}
- {teammate feedback from the retro conversation}

### Recommendations
- {what to change next time — e.g., "start with scout driving orientation before craftsman drives implementation"}
- {what to keep — e.g., "the expert-as-navigator pattern caught 2 real bugs"}

This file is for the human. Read it before starting the next popcorn-xp session on the same codebase — it's the team's institutional memory about how the process works here.

Advice System

Strong opinions, loosely held. The driver has their own approach and should defend it. Advice is input from a different lens — not instructions to follow. The navigator might be wrong. The driver might be wrong. The point is engagement, not compliance.

TypeMeaningDriver responseBlocks?
OBJECTION"This is wrong."Engage: fix if they're right, reject with reasoning if not. Both valid.Yes
SMELL"This looks off."Read it. Use your judgment. Acknowledge when you can.No
STEER"Try a different approach."Consider it honestly. Your way might be better.No
FYI"Noticed this, might matter later."Noted.No

Only OBJECTIONs block task completion. Everything else is the driver's call. The navigator should use OBJECTION sparingly — overusing it makes them a blocker, not a partner.

Advice is sent via SendMessage (real-time) and appended to .popcorn-xp/ADVICE.md (persistent record).

Three hooks support the advice lifecycle:

  • TaskCompleted — blocks on open OBJECTIONs, reminds of other open advice
  • TeammateIdle — reminds the agent of open advice items when they go idle
  • SubagentStop — backup block on open OBJECTIONs

Session Files

Three files in .popcorn-xp/:

  • LOG.md — Append-only execution history. What was done, what was learned, what was decided. Written by teammates during the session.
  • ADVICE.md — Persistent record of all typed advice and resolutions. Written by teammates during the session.
  • RETRO.md — Accumulated retrospective entries. Written by the lead after each session. Read this before starting the next session — it's the team's process memory.

The first teammate to start work creates .popcorn-xp/, LOG.md, and ADVICE.md. The lead creates RETRO.md after TeamDelete.

Quality Bar

  • Only one driver edits at a time. The navigator reads and advises. No concurrent edits.
  • OBJECTIONs block task completion. No exceptions.
  • Different roles must contribute materially different perspectives.
  • Task descriptions carry enough context for independent execution.
  • LOG.md is detailed enough that the next agent can reconstruct what happened.
  • The lead manages the team but does not do the work.
  • Stop spawning rounds when additional tasks stop changing the plan.

Reference

Read references/protocol.md for teammate prompt templates, advice format, and session file templates. Include the relevant protocol sections in teammate prompts when spawning them.

平台分布

Codex

34.49%
按下载量换算30

Claude

29.94%
按下载量换算26

Cursor

20.6%
按下载量换算18

Gemini CLI

9.86%
按下载量换算8

安全审计

Gen Agent Trust Hub

通过

Socket

可疑

Snyk

通过

权限和风险

执行命令

安装流程涉及命令执行,可能通过 npx skills add https://github.com/mikewolfd/xp-popcorn-skill --skill popcorn-xp 联网下载 Skill 或依赖。用户安装前应确认命令来源、仓库内容和执行环境。

安装前确认

本站仅展示第三方公开信息,不托管安装包,不提供自动安装或运行环境。安装前应自行审查源码、依赖和命令行为。来源安全扫描存在提醒或未通过项,安装前请自行复核。目前只有一个公开来源,建议同时核对源仓库。

继续浏览同类 Skills