원클릭으로 Manus에서 모든 스킬 실행

browser-mcp-resource-guard

스타12

포크0

업데이트2026년 7월 2일 07:30

Use when Codex is about to use, is using, or has just used browser automation MCP tools such as chrome-devtools-mcp, Chrome DevTools MCP, Playwright MCP, @playwright/mcp, browser MCP, or browser/computer automation helpers; when the user mentions high CPU, high energy usage, battery drain, fans, heat, lag, duplicate browser MCP processes, stale Chrome Helper processes, or MCP cleanup; or when long-running Codex sessions may leave browser automation MCP helper processes behind.

설치

Codex 또는 Claude로 설치 이 Prompt를 복사해 Codex, Claude 또는 다른 어시스턴트에 붙여 넣으면 Skill 페이지를 검토하고 설치를 진행할 수 있습니다.

Manus에서 실행

출처

nangongwentian-fe

nangongwentian-fe/jay-skills

GitHub 저장소 열기 Creator 저장소 보기

다운로드

Manus에서 실행

Browser MCP Resource Guard

Goal

Reduce CPU, memory, battery, and fan impact from browser automation MCP tools without breaking active work. Treat cleanup as a guarded operation: inspect first, report clearly, and only kill processes when it is safe or explicitly requested.

Decision Rules

Prefer not using browser MCP when static checks, local files, tests, builds, or ordinary HTTP requests answer the task.
Use Chrome DevTools MCP when the task needs live browser state, console messages, network requests, DOM snapshots, screenshots, viewport checks, or real interactions.
Use browser MCP in focused batches: open the page, collect needed evidence, close or stop related local services when finished.
Avoid repeated screenshots, repeated new tabs, performance traces, heap snapshots, and attaching to a Chrome profile with many tabs unless the task requires them.
After browser MCP use, run a read-only process check when the session was long, multiple pages/tools were used, or the user mentions energy/resource problems.

Read-Only Check

Run the bundled report script from the skill directory:

scripts/browser_mcp_process_report.sh

If the script is unavailable, use:

ps -axo pid=,ppid=,rss=,etime=,command= \
  | rg 'chrome-devtools-mcp|@playwright/mcp|playwright-mcp|ms-playwright|mcp-chrome'

Report:

number of browser MCP-related process rows
rough RSS total
longest elapsed time
whether there are duplicate stacks
whether any process appears related to the current task

Cleanup Policy

Do not blindly kill all MCP processes.
Ask before killing when a process may belong to another active thread, browser task, IDE session, or user workflow.
It is usually safe to propose cleanup when there are multiple duplicate chrome-devtools-mcp or Playwright MCP stacks, old elapsed times, no current browser automation work, and the user is reporting energy or lag.
Prefer terminating only browser MCP helper processes, not the user's normal Chrome profile.

Use these commands only after approval or when the user explicitly asks to clean clear browser MCP leftovers:

pkill -f 'chrome-devtools-mcp'
pkill -f '@playwright/mcp'
pkill -f 'playwright-mcp'

For a current-turn dev server started by Codex, stop the specific shell session or process instead of using broad pkill.

Final Response

State the result in one or two short sections:

what was found, with process count/RSS if checked
what was cleaned or what should be cleaned
any remaining risk, such as possible active browser automation in another thread

이 저장소의 다른 Skills

같은 저장소

daily-work-summary

nangongwentian-fe/jay-skills

根据指定 Git 仓库和日期，从 commit、提交正文、文件变更及必要的 diff 中提炼每日工作内容。用于用户要求按项目和日期生成日报、每日工作总结、根据 Git commit 总结工作，或要求用 1、2、3 编号列出某天或某段日期完成事项的场景。

2026-07-1812

web-content-fetcher

nangongwentian-fe/jay-skills

获取公开网页的正文或 Markdown 内容。当用户要求抓取链接、读取文章、提取网页正文、将页面转为 Markdown，或获取 X/Twitter、微信、知乎、Medium 等公开页面的文字内容时使用。支持在直接抓取失败或内容不完整时，通过内容提取代理和网页搜索逐级回退；不适用于需要登录的页面、付费墙绕过、媒体下载或结构化数据抓取。

2026-07-1312

git-rebase-workflow

nangongwentian-fe/jay-skills

Git Rebase 分支同步流程，用于将当前功能分支 rebase 到最新的目标分支（如 master/main），保持提交历史整洁。适用于功能分支落后于目标分支时，需要同步最新代码的场景。

2026-07-0512

post-task-learning-review

nangongwentian-fe/jay-skills

Review a completed or nearly completed non-trivial task and maintain durable lessons across project docs, memory, existing skills, or new skills. Use after complex debugging, deployment, live verification, repo documentation work, repeated workflow discovery, creating or updating a skill, or when the user asks whether anything is worth remembering, updating, or deleting.

2026-07-0512

git-topic-commit-push

nangongwentian-fe/jay-skills

Create one or more Git commits grouped by coherent change topic, then push the current branch. Use when the user asks to commit and push, submit by topic, split current changes into topical commits, or do "按照主题提交 commit 并 push"; especially when a worktree has mixed staged, unstaged, or untracked changes that need honest commit boundaries before `git push`. Commit messages must be written in Chinese unless the user explicitly requests another language.

2026-07-0112

goal-loop-builder

nangongwentian-fe/jay-skills

Create copy-ready, verifiable run contracts for long-horizon agent work. Use when the user asks to write, improve, or review Codex `/goal` prompts, Claude Code `/goal` conditions, Claude `/loop` prompts, `.claude/loop.md`, `goal.md`, loop engineering prompts, autonomous run instructions, stop/pause conditions, verification criteria, bounded iteration policy, or persistent Markdown instructions for agents.

2026-06-1612

name

browser-mcp-resource-guard

description

Browser MCP Resource Guard

Goal

Decision Rules

Prefer not using browser MCP when static checks, local files, tests, builds, or ordinary HTTP requests answer the task.
Use Chrome DevTools MCP when the task needs live browser state, console messages, network requests, DOM snapshots, screenshots, viewport checks, or real interactions.
Use browser MCP in focused batches: open the page, collect needed evidence, close or stop related local services when finished.
Avoid repeated screenshots, repeated new tabs, performance traces, heap snapshots, and attaching to a Chrome profile with many tabs unless the task requires them.
After browser MCP use, run a read-only process check when the session was long, multiple pages/tools were used, or the user mentions energy/resource problems.

Read-Only Check

Run the bundled report script from the skill directory:

scripts/browser_mcp_process_report.sh

If the script is unavailable, use:

ps -axo pid=,ppid=,rss=,etime=,command= \
  | rg 'chrome-devtools-mcp|@playwright/mcp|playwright-mcp|ms-playwright|mcp-chrome'

Report:

number of browser MCP-related process rows
rough RSS total
longest elapsed time
whether there are duplicate stacks
whether any process appears related to the current task

Cleanup Policy

Do not blindly kill all MCP processes.
Ask before killing when a process may belong to another active thread, browser task, IDE session, or user workflow.
It is usually safe to propose cleanup when there are multiple duplicate chrome-devtools-mcp or Playwright MCP stacks, old elapsed times, no current browser automation work, and the user is reporting energy or lag.
Prefer terminating only browser MCP helper processes, not the user's normal Chrome profile.

Use these commands only after approval or when the user explicitly asks to clean clear browser MCP leftovers:

pkill -f 'chrome-devtools-mcp'
pkill -f '@playwright/mcp'
pkill -f 'playwright-mcp'

For a current-turn dev server started by Codex, stop the specific shell session or process instead of using broad pkill.

Final Response

State the result in one or two short sections:

what was found, with process count/RSS if checked
what was cleaned or what should be cleaned
any remaining risk, such as possible active browser automation in another thread