Token导航 LogoToken导航

Agent Skill · 研究检索

ditch-vercel(ditch Vercel 搜索)

ditch-vercel 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。

ditch-vercel 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,权限为接触敏感数据;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 72,323。

总安装 188周安装 8GitHub Stars 1下载 66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:ditch-vercel(ditch Vercel 搜索)
来源仓库:https://github.com/umarmuhandis/ditch-vercel
仓库路径:skills/ditch-vercel
安装命令:
npx skills add https://github.com/umarmuhandis/ditch-vercel --skill ditch-vercel
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/umarmuhandis/ditch-vercel --skill ditch-vercel

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。

ditch-vercel 提供从 Vercel 迁移至自建部署环境的五阶段操作指南。适用于需要将 Next.js 等项目从 Serverless 平台迁移至自有基础设施的场景。按顺序执行环境评估、数据导出、配置迁移与验证测试。禁止跳过阶段或提前修改文件,需用户逐阶段确认后推进。

适用宿主包括 Codex、Claude、Cursor、Gemini CLI,接入前应确认版本、权限和运行环境要求。

来源仓库内路径:skills/ditch-vercel

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/umarmuhandis/ditch-vercel --skill ditch-vercel;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

    它需要接触敏感数据,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

ditch-vercel 怎么安装?

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

ditch-vercel 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。

ditch-vercel 需要什么权限?

权限级别是接触敏感数据。该 Skill 可能接触密钥、Token、环境变量或敏感配置,应进入高风险复核队列,默认不自动发布。

ditch-vercel 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。

ditch-vercel 用的人多吗?

公开数据是累计安装 188 次,近一周安装 8 次,下载 66 次,GitHub 1 星。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。

ditch-vercel 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

  • 需要接触敏感数据,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

  • 环境里有大量密钥、凭证,不方便给第三方 Skill 读取的人。
  • 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
ditch-vercel

ditch Vercel 搜索 · 本页

1881接触敏感数据未明确有提醒

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

Ditch Vercel — Migration Orchestrator

You are running the ditch-vercel migration skill. Follow the 5-phase flow below exactly in order. Do NOT skip phases. Do NOT make any file changes until Phase 4 (after explicit user approval in Phase 3).


Visual Style

All user-facing output must use the cyberpunk "Escape Sequence" visual language. Follow these rules when rendering output in every phase:

  1. Major frames (Phase 1 banner, Phase 5 completion): Use double-line box-drawing characters — ╔ ═ ╗ ║ ╚ ═ ╝
  2. Reports and sub-panels (Phase 2 report, Phase 3 plan): Use single-line rounded box-drawing — ┌ ─ ┐ │ └ ─ ┘
  3. Narrative/status lines: Prefix with > — e.g. > VERCEL LOCK-IN DETECTED
  4. Progress bars (Phase 4): ██ for filled segments, ░░ for empty segments
  5. Status tags (Phase 4 tracker): CLEAR (completed), ACTIVE (in progress), QUEUED (pending), FAILED (error), SKIPPED (user skipped)
  6. Prefix symbols used across phases:

- [✓] — done / automated - [!] — needs attention - [✗] — blocker / critical - + — add - - — remove - ~ — modify - × — delete - ◆ — manual action


Phase 1: SCAN

Silently detect framework, Vercel features, and target platform in one pass. Minimal output — the report comes in Phase 2.

Pre-flight check

Verify the project is a git repository by checking for a .git directory. If not a git repo, warn the user: "This project is not a git repository. The safety checkpoint (Phase 4) requires git for rollback. Initialize with git init first, or proceed without rollback protection." Ask the user whether to continue or stop.

1a. Detect framework

Read package.json — examine dependencies and devDependencies. Check for framework config files. Match against this table (first match wins):

Frameworkpackage.json indicatorConfig file pattern
Next.jsnext in depsnext.config.*
Astroastro in depsastro.config.*
Remix@remix-run/* in depsremix.config.* (optional)
SvelteKit@sveltejs/kit in depssvelte.config.*
Nuxtnuxt in depsnuxt.config.*
StaticNone of the above framework depsN/A

Read the corresponding knowledge file:

1b. Detect ALL Vercel features

Scan for every Vercel-specific feature. Check every item — do NOT skip any.

Configuration:

  • vercel.json — Read it fully if present. Note: rewrites, redirects, headers, cron, functions config, regions.

Vercel SDK packages — Check package.json deps and devDeps for:

  • @vercel/analytics
  • @vercel/speed-insights
  • @vercel/blob
  • @vercel/kv
  • @vercel/postgres
  • @vercel/edge
  • @vercel/og
  • @vercel/edge-config
  • Any other @vercel/* package

Framework-specific Vercel features — Scan source files:

  • Edge Runtime: Search for export const runtime = 'edge' or export const runtime = "edge" in route/API files.
  • next/image: Search for import.*from ['"]next/image['"] — Vercel's image optimization is used.
  • ISR: Search for export const revalidate or revalidate: in page/route files.
  • API Routes: Search for app/api/**/route.{ts,js} or pages/api/**/*.{ts,js} (Next.js), or equivalent in other frameworks.
  • Middleware: Check for middleware.{ts,js} at project root or src/.
  • Cron jobs: Check vercel.json for crons field.
  • Environment variables: Check for .env* files. Note which env vars exist (names only, never values).
  • Serverless/Edge functions config: Check for export const config = {runtime:...} patterns.

1c. Select target platform

Ask the user to choose the target platform:

Where do you want to migrate?
- Cloudflare (Workers/Pages — serverless edge) (Recommended)
- Railway (Node.js — managed infrastructure with native DBs)
- VPS (Node.js + PM2 + Nginx — self-managed server)
- Other targets coming soon (Netlify, Fly.io)

Read the target knowledge file:

1d. Output

Display the opening banner, then the themed scan results:

╔══════════════════════════════════════════════╗
║  ██████╗ ██╗████████╗ ██████╗██╗  ██╗       ║
║  ██╔══██╗██║╚══██╔══╝██╔════╝██║  ██║       ║
║  ██║  ██║██║   ██║   ██║     ███████║       ║
║  ██║  ██║██║   ██║   ██║     ██╔══██║       ║
║  ██████╔╝██║   ██║   ╚██████╗██║  ██║       ║
║  ╚═════╝ ╚═╝   ╚═╝    ╚═════╝╚═╝  ╚═╝       ║
║                  VERCEL                       ║
╚══════════════════════════════════════════════╝

> VERCEL LOCK-IN DETECTED
> INITIATING ESCAPE SEQUENCE...

  Framework lock identified :  [Framework] [version] ([variant])
  Hostile packages found    :  [N] Vercel-specific features
  Escape route              :  [Target platform]

Example values: Next.js 15 (App Router), 8, Cloudflare Workers/Pages


Phase 2: REPORT

The anxiety reducer. Show the developer exactly how hard this migration is BEFORE asking them to commit to anything.

2a. Calculate complexity score

Cross-reference every detected Vercel feature against the target's compatibility matrix (from the target knowledge file) and the framework knowledge file's compatibility notes.

Deduplication rule: When the same feature appears in BOTH the target matrix and the framework compatibility notes:

  1. Use the target file's Weight as the base score
  2. If the framework file lists a lower weight for the same feature on the same target, use the lower weight (the framework's adapter may handle it automatically)
  3. Count each feature exactly once — never sum from both files

For each detected feature:

  1. Look up its Weight and Category — target file is primary, framework file can override downward only
  2. Sum all weights → total complexity score
  3. Determine the traffic light:

- 🟢 GREEN (0-2): ~1-2 hours — mostly automated - 🟡 YELLOW (3-6): ~3-5 hours — several manual steps - 🔴 RED (7+): ~1-2 days — significant refactoring or blockers

2b. Categorize features into 3 groups

  • Automated (Weight 0): ditch-vercel handles these entirely
  • Attention (Weight 1): Works but needs minor manual adjustment
  • Blocker (Weight 3): Significant effort, may prevent migration

2c. Output the report

> THREAT ASSESSMENT COMPLETE

┌──────────────────────────────────────────────┐
│  LOCK-IN SEVERITY : [🟢 GREEN / 🟡 YELLOW / 🔴 RED]  │
│  Estimated extraction time : [time estimate]  │
└──────────────────────────────────────────────┘

  NEUTRALIZED (auto-handled):
  [✓] [feature] → [what happens]
  [✓] [feature] → [what happens]
  ...

  REQUIRES OPERATOR INTERVENTION:
  [!] [feature] → [what the developer needs to do]
  [!] [feature] → [what the developer needs to do]
  ...

  CRITICAL THREATS:
  [✗] [feature] → [why it's a blocker and what's needed]
  ...

Only include sections that have items. If there are no Critical Threats, omit that section. If there are no Operator Intervention items, omit that section.


Phase 3: PLAN + APPROVE

Generate a concrete migration plan, get explicit approval, then create the task list.

3-pre. Cross-reference official docs

Before generating the plan, verify migration steps against current official documentation.

  1. Read the ## Reference URLs section from the loaded framework file and target file
  2. Prefer llms.txt URLs (lines prefixed with llms.txt:) — these are machine-readable doc indexes optimized for LLMs. Fetch the llms.txt URL and use the index to locate the specific migration/deployment page, then fetch that page. If no llms.txt entry exists for a source, fall back to the regular doc URLs.
  3. Fetch each URL and extract: current migration steps, required packages, config format, and any breaking changes or deprecation notices
  4. Compare fetched content against the migration steps in the framework knowledge file. Look for:

- Package name changes (e.g. adapter renamed) - New required config fields - Deprecated CLI flags or commands - Changed build output directories

  1. If discrepancies found: note them. They will be incorporated into the plan in 3a and flagged for the user in the plan output.
  2. If fetching fails for any URL: skip silently, proceed with framework file instructions

3a. Generate the migration plan

Based on Phases 1-2 and the doc verification in 3-pre, produce a specific plan. List exact file paths, package names, and what changes will be made. Tag each item with its category. If doc verification found discrepancies with the knowledge file, use the official docs as the source of truth and note the discrepancy in the relevant plan item.

> EXTRACTION PLAN GENERATED
══════════════════════════════════════════════════
  [Framework] on Vercel → [Target]
══════════════════════════════════════════════════

  PAYLOADS TO DEPLOY:
  + [package]  [AUTO]
  + [package]  [AUTO]
  ...

  PAYLOADS TO JETTISON:
  - [package]  [AUTO]
  - [package]  [AUTO]
  ...

  FILES TO INJECT:
  + [filepath] — [description]  [AUTO]
  ...

  FILES TO PATCH:
  ~ [filepath] — [description]  [AUTO/MANUAL]
  ...

  FILES TO PURGE:
  × [filepath]  [AUTO]
  ...

  POST-EXTRACTION OPS:
  ◆ [item]  [MANUAL]
  ...

Each entry must be specific enough that the developer understands exactly what will happen. Tag each item [AUTO] for automated or [MANUAL] for attention/blocker items.

3b. Approval gate

This is a hard gate. Do NOT proceed without explicit approval.

Ask the user:

> AUTHORIZE EXTRACTION?

  [1] Yes — execute extraction sequence
  [2] Modify — adjust the plan
  [3] Abort — no changes, exit clean

If "Modify": loop — gather feedback, revise the plan, present for approval again. If "Abort": stop gracefully. Output: > EXTRACTION ABORTED. No files changed. Run /ditch-vercel anytime to re-engage.

CRITICAL: Do NOT create, modify, or delete any project files before receiving approval (option 1).

3c. Build the execution checklist

After the developer approves, build an ordered checklist of every migration action. Track each item's progress throughout Phase 4 (announce when starting and completing each step). If your agent supports built-in task tracking, use it.

Execution order:

  1. Create git safety checkpoint
  2. One item per dependency to install
  3. One item per dependency to remove
  4. One item per file to create
  5. One item per file to modify
  6. One item per file to delete
  7. Run build verification
  8. Run local dev server verification

Each item must describe one atomic action with the exact command or file change.

Phase 4: EXECUTE

Execute the approved plan with safety nets and real-time task tracking.

Output the extraction authorization header:

> EXTRACTION SEQUENCE AUTHORIZED
──────────────────────────────────────────────

4a. Git safety checkpoint

Begin the git safety checkpoint.

  1. Check git status. If working tree is dirty: git add -u && git commit -m "chore: pre-migration checkpoint (ditch-vercel)" Warning: Only stage tracked files (git add -u). Do NOT use git add -A — it can accidentally commit .env files or credentials. If there are important untracked files the user wants to preserve, tell them to git add those specific files first.
  2. If working tree is clean, note the current HEAD SHA.
  3. Store the checkpoint SHA for rollback.
  4. Mark the checkpoint step as done.
  5. Output: > Safety checkpoint locked (commit: <sha-short>). Emergency rollback: git reset --hard <sha>

4b. Detect package manager

Check the project root for lock files (first match wins):

Lock filePackage manager
bun.lockb or bun.lockbun
pnpm-lock.yamlpnpm
yarn.lockyarn
package-lock.jsonnpm

If no lock file found, default to npm.

4c. Execute each task

For each remaining task (dependencies, files, deletions):

Important: Follow the migration steps section in the framework knowledge file that matches the selected target platform. For example, if the target is VPS, follow ## Migration Steps (VPS). If the target is Cloudflare, follow ## Migration Steps (Cloudflare). Use the corresponding ## Compatibility Notes ([target]) section for replacement guidance.

  1. Execute the action:

- Install dep: [pkg-manager] add <pkg> / [pkg-manager] add -D <pkg> - Remove dep: [pkg-manager] remove <pkg> - Create file: Create the file - Modify file: Edit the file - Delete file: Delete the file (e.g. rm)

  1. If successful: mark the step as done
  2. If failed: show the error and ask the developer:

- "Fix it" → Read the error, attempt a fix, retry the action - "Skip this step" → Mark done with SKIPPED, continue - "Rollback everything" → Run git reset --hard <checkpoint-sha>, skip all remaining steps, stop execution

Progress tracker rendering: After each step completes, re-render the full progress tracker showing all steps. Use this format:

[03/14] ██████████████████░░░░░░░░░░░░  ACTIVE  Installing @opennextjs/cloudflare
───────────────────────────────────────────────
  [01/14] ██████████  CLEAR    Git safety checkpoint
  [02/14] ██████████  CLEAR    Remove @vercel/analytics
  [03/14] █████░░░░░  ACTIVE   Installing @opennextjs/cloudflare
  [04/14] ░░░░░░░░░░  QUEUED   Create wrangler.toml
  ...

Status tags: CLEAR (completed), ACTIVE (in progress), QUEUED (pending), FAILED (error), SKIPPED (user skipped). The top line shows the current step; the list shows all steps.

On failure after the user chooses rollback, output:

> ABORT EXTRACTION? Reset to checkpoint <sha>

4d. Build verification

After all file changes are complete:

  1. Begin build verification
  2. Detect the build command:

- Next.js + Cloudflare: <pkg> run build:cf (or the build:cf script added in migration) - Next.js + VPS: <pkg> run build - Astro: <pkg> run build (runs astro build) - Remix: <pkg> run build (runs remix vite:build) - SvelteKit: <pkg> run build - Nuxt: <pkg> run build (runs nuxt build) - Next.js + Railway: <pkg> run build - Astro + Railway: <pkg> run build - Remix + Railway: <pkg> run build - SvelteKit + Railway: <pkg> run build - Nuxt + Railway: <pkg> run build - Static + Railway: <pkg> run build (or skip if no build step) - Static with build script: <pkg> run build - Static without build: skip (no build needed)

  1. Run the build command
  2. If build passes: mark step as done
  3. If build fails:

- Show the error output (first 50 lines) - Ask the developer: - "Fix it" → Read the error, attempt to fix the code, re-run the build - "Rollback everything" → Run git reset --hard <checkpoint-sha>, skip remaining steps, stop - "Continue anyway" → Note "[BUILD FAILED - manual fix needed]" and continue

4e. Local dev server verification

After build passes, verify the app starts and responds locally.

  1. Begin local dev server verification
  2. Detect the preview command:

- Next.js + Cloudflare: npx wrangler dev (port 8787) - Next.js + VPS: node.next/standalone/server.js (port 3000) - Astro + Cloudflare: npx wrangler pages dev dist/ (port 8788) - Astro + VPS: node dist/server/entry.mjs (port 4321) - Remix + Cloudflare: npx wrangler pages dev build/client (port 8788) - Remix + VPS: npx remix-serve build/server/index.js (port 3000) - SvelteKit + Cloudflare: npx wrangler pages dev.svelte-kit/cloudflare (port 8788) - SvelteKit + VPS: node build/index.js (port 3000) - Nuxt + Cloudflare: npx wrangler pages dev.output/public (port 8788) - Nuxt + VPS: node.output/server/index.mjs (port 3000) - Static + Cloudflare: npx wrangler pages dev <output-dir> (port 8788) - Next.js + Railway: node.next/standalone/server.js (port 3000) - Astro + Railway: node dist/server/entry.mjs (port 4321) - Remix + Railway: npx remix-serve build/server/index.js (port 3000) - SvelteKit + Railway: node build/index.js (port 3000) - Nuxt + Railway: node.output/server/index.mjs (port 3000) - Static + Railway: npx serve <output-dir> (port 3000) - Static + VPS: npx serve <output-dir> (port 3000)

  1. Start the preview command in the background
  2. Wait ~5 seconds for the server to start
  3. Run curl -s -o /dev/null -w "%{http_code}" http://localhost:<port>/ to check for a 200 response
  4. Kill the background process
  5. If curl returns 200: mark step as done
  6. If curl fails or non-200:

- Show the output - Ask the developer: - "Fix it" → investigate, fix, retry - "Skip" → mark done with "[DEV SERVER CHECK SKIPPED]" - "Rollback everything" → Run git reset --hard <checkpoint-sha>, skip remaining steps, stop

Phase 5: DONE

Output the final migration summary with everything the developer needs.

5a. Migration summary

╔══════════════════════════════════════════════╗
║  ███████╗██████╗ ███████╗███████╗            ║
║  ██╔════╝██╔══██╗██╔════╝██╔════╝            ║
║  █████╗  ██████╔╝█████╗  █████╗              ║
║  ██╔══╝  ██╔══██╗██╔══╝  ██╔══╝              ║
║  ██║     ██║  ██║███████╗███████╗            ║
║  ╚═╝     ╚═╝  ╚═╝╚══════╝╚══════╝            ║
╚══════════════════════════════════════════════╝

> EXTRACTION COMPLETE
> You are no longer locked in.

  Created  :  [list files]
  Patched  :  [list files]
  Purged   :  [list files]
  Deployed :  [list packages]
  Removed  :  [list packages]

  REMAINING OPS:
  ◆ [item from report that wasn't fully automated]
  ...

  NEXT MOVES:
  1. [deploy command] — deploy to [target]
  2. Set environment variables in [target] dashboard
  3. [any other manual items]

  EMERGENCY ROLLBACK:
  git reset --hard [checkpoint-sha]

Only include "REMAINING OPS" if there are remaining items. Derive the local dev command, deploy command, and manual items from the target knowledge file and framework knowledge file.


General Rules

  • Be precise: Use exact file paths, package names, and commands. Never be vague.
  • Be safe: Never touch files outside the project directory. Never expose env var values.
  • Be transparent: Show the user what you found and what you plan to do before doing it.
  • Handle errors: If a knowledge file is missing, work from your own knowledge but warn the user. If a scan finds nothing, say so explicitly rather than guessing.
  • One framework only: If multiple frameworks are detected, ask the user to clarify which one is the primary framework.
  • Monorepo awareness: If the project root contains apps/ or packages/ directories, ask the user which app to migrate.
  • Track progress through each step during Phase 4. Use your agent's task tracking if available.
  • Step #1 must always be the git safety checkpoint.
  • Never mark a step as done if the action failed. Failed steps need recovery options presented to the developer. Note: user-initiated skips (where the developer explicitly chooses "Skip this step") are NOT failures — mark these done with a "[SKIPPED]" tag.

平台分布

Codex

35.83%
按下载量换算24

Claude

31.74%
按下载量换算21

Cursor

18.17%
按下载量换算12

Gemini CLI

8.84%
按下载量换算6

安全审计

Gen Agent Trust Hub

通过

Socket

可疑

Snyk

可疑

权限和风险

接触敏感数据

该 Skill 可能接触密钥、Token、环境变量或敏感配置,应进入高风险复核队列,默认不自动发布。

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