skills/install-recommended-tools/SKILL.md
Walk through installing recommended Claude Code plugins, LSPs, and MCP servers based on your development type and languages.
npx skillsauth add agentic-cookbook/tools install-recommended-toolsInstall this skill globally with one command. Works with Claude Code, Cursor, and Windsurf.
3 of 9 scanners reported clean
Some scanners were skipped, did not run, or reported a non-clean status. Review each row below.
First action: If $ARGUMENTS is --version, print install-recommended-tools v2.0.0 and stop.
Otherwise, print install-recommended-tools v2.0.0 as the first line of output, then proceed.
An interactive guide that recommends Claude Code plugins, LSP integrations, and MCP servers based on what you build and the languages you use. It detects what's already installed and only recommends what's missing.
Each recommendation specifies a scope — global or local:
The recommendations come from the developer-tools research catalog at research/developer-tools/ in the agentic-cookbook repo.
These plugins register hooks that fire on every tool call or prompt. Never install them globally — the hooks run in every project whether relevant or not.
| Plugin | Hooks | Fires on |
|--------|-------|----------|
| semgrep | PostToolUse, UserPromptSubmit, SessionStart (3) | Every write + every prompt |
| security-guidance | PreToolUse (1) | Every Write/Edit |
| hookify | PreToolUse, PostToolUse, UserPromptSubmit, Stop (4) | Everything |
| vercel | 13 hooks across all events | Everything + telemetry |
| railway | PreToolUse (1) | Every Bash command |
When recommending any of these, always set scope to local and include the warning: "Has hooks that fire continuously — install per-project only."
Read these files to build the installed inventory:
~/.claude/settings.json — extract the enabledPlugins array. Each entry is <plugin-name>@<marketplace>. If the file doesn't exist or is malformed, treat the list as empty..mcp.json in the project root (if it exists) — extract keys from mcpServers. If missing, treat as empty.~/.claude.json (if it exists) — extract MCP servers configured at user scope. If missing, treat as empty.Build two lists:
installed_plugins: plugin names (without marketplace suffix)installed_mcp: MCP server namesPrint a compact summary:
=== Installed Tools ===
Plugins: 33 installed (typescript-lsp, swift-lsp, semgrep, superpowers, ...)
MCP servers: 1 configured (playwright)
Use AskUserQuestion to ask:
What types of development do you work on? (select all that apply)
- Web frontend (React, Vue, Svelte, etc.)
- Web backend (APIs, servers, databases)
- iOS / macOS (Swift, Xcode)
- Android (Kotlin, Gradle)
- Windows / .NET (C#, WinUI)
- CLI / terminal apps
- Data / ML pipelines
- Infrastructure / DevOps
- API development (REST, GraphQL, gRPC)
Accept comma-separated numbers (e.g., "1, 2, 9"). If the user provides invalid input (non-numeric, out of range), ask again with a clarifying note.
Use AskUserQuestion to ask:
What languages do you primarily use? (select all that apply)
- TypeScript / JavaScript
- Python
- Swift
- Kotlin
- C# / .NET
- Go
- Rust
- Java
- Ruby
- PHP
- C / C++
- Elixir
- Lua
Accept comma-separated numbers.
Assemble a recommendation list based on selections. Each recommendation has a tier (Essential, Recommended, or Optional), a one-line description, and an install command.
Skip any tool already in installed_plugins or installed_mcp.
| Tool | Tier | Scope | Description |
|------|------|-------|-------------|
| superpowers | Essential | global | TDD, systematic debugging, brainstorming, subagent workflows |
| code-review | Essential | global | Multi-agent PR code review with confidence filtering |
| pr-review-toolkit | Essential | global | Specialized review agents across 6 domains |
| context7 | Essential | global | Real, version-specific library docs in context |
| github | Essential | global | GitHub API: issues, PRs, code search |
| security-guidance | Recommended | local | Scans file writes for 8 vulnerability categories. Has PreToolUse hook — install per-project only. |
| semgrep | Recommended | local | Real-time SAST scanning via MCP server. Has 3 continuous hooks — install per-project only. Requires SEMGREP_APP_TOKEN. |
| claude-code-setup | Recommended | global | Analyze codebase and recommend automations |
| commit-commands | Optional | global | Git commit, push, PR commands |
| claude-md-management | Optional | global | Maintain and audit CLAUDE.md files |
| feature-dev | Optional | global | 7-phase multi-agent feature development |
| remember | Optional | global | Continuous cross-session memory via daily logs |
Map each selected language to its LSP plugin:
| Language | Plugin | Binary Required | Install |
|----------|--------|-----------------|---------|
| TypeScript/JavaScript | typescript-lsp | typescript-language-server | /plugin install typescript-lsp@claude-plugins-official |
| Python | pyright-lsp | pyright-langserver | /plugin install pyright-lsp@claude-plugins-official |
| Swift | swift-lsp | sourcekit-lsp | /plugin install swift-lsp@claude-plugins-official |
| Kotlin | kotlin-lsp | kotlin-language-server | /plugin install kotlin-lsp@claude-plugins-official |
| C# / .NET | csharp-lsp | csharp-ls | /plugin install csharp-lsp@claude-plugins-official |
| Go | gopls-lsp | gopls | /plugin install gopls-lsp@claude-plugins-official |
| Rust | rust-analyzer-lsp | rust-analyzer | /plugin install rust-analyzer-lsp@claude-plugins-official |
| Java | jdtls-lsp | jdtls | /plugin install jdtls-lsp@claude-plugins-official |
| Ruby | ruby-lsp | ruby-lsp | /plugin install ruby-lsp@claude-plugins-official |
| PHP | php-lsp | intelephense | /plugin install php-lsp@claude-plugins-official |
| C / C++ | clangd-lsp | clangd | /plugin install clangd-lsp@claude-plugins-official |
| Elixir | elixir-ls-lsp | elixir-ls | /plugin install elixir-ls-lsp@claude-plugins-official |
| Lua | lua-lsp | lua-language-server | /plugin install lua-lsp@claude-plugins-official |
All LSP plugins are Essential tier for their language and always local scope (language-specific).
All dev-type-specific plugins are local scope unless noted otherwise.
Web frontend (selection 1):
| Tool | Tier | Description |
|------|------|-------------|
| frontend-design | Essential | Production-grade UI with distinctive design quality |
| figma | Essential | Read Figma designs for design-to-code translation |
| playwright | Essential | Browser automation, E2E testing, screenshots |
| playground | Recommended | Interactive HTML playgrounds with live preview |
| accesslint | Recommended | WCAG accessibility compliance checking |
| stagehand | Optional | Natural language browser automation |
Web backend (selection 2):
| Tool | Tier | Description |
|------|------|-------------|
| postman | Recommended | Full API lifecycle: collections, tests, mocks, docs |
| sentry | Optional | Error monitoring, stack traces, issue search |
| prisma | Optional | Prisma ORM, migrations, PostgreSQL management |
iOS / macOS (selection 3):
| Tool | Tier | Description |
|------|------|-------------|
| figma | Recommended | Design-to-code for iOS UI |
Android (selection 4):
| Tool | Tier | Description |
|------|------|-------------|
| figma | Recommended | Design-to-code for Android UI |
Windows / .NET (selection 5):
| Tool | Tier | Description |
|------|------|-------------|
| microsoft-docs | Recommended | Azure and .NET documentation lookup |
CLI / terminal (selection 6):
| Tool | Tier | Description |
|------|------|-------------|
| ralph-loop | Recommended | Autonomous multi-hour coding sessions. Has Stop hook. |
Data / ML (selection 7):
| Tool | Tier | Description |
|------|------|-------------|
| huggingface-skills | Recommended | Build and train ML models with Hugging Face |
Infrastructure / DevOps (selection 8):
| Tool | Tier | Description |
|------|------|-------------|
| deploy-on-aws | Recommended | 5-step AWS deployment with cost estimation |
| terraform | Recommended | Terraform IaC automation |
| vercel | Optional | Vercel deployment. Has 13 continuous hooks — install per-project only. |
| railway | Optional | Railway container deployment. Has PreToolUse hook on every Bash — install per-project only. |
| netlify-skills | Optional | Netlify serverless/edge/databases |
API development (selection 9):
| Tool | Tier | Description |
|------|------|-------------|
| postman | Essential | API collections, testing, mocks, documentation |
Deduplicate the combined list (a tool recommended by multiple dev types appears once, at its highest tier).
Group by scope first, then by tier within each scope:
=== Recommended Tools ===
GLOBAL (install once, available in every project)
Essential:
1. superpowers — TDD, debugging, brainstorming, subagent workflows
2. code-review — Multi-agent PR code review
...
Recommended:
5. claude-code-setup — Analyze codebase and recommend automations
...
LOCAL (install in this project only)
Essential:
7. typescript-lsp — TypeScript/JS type checking and diagnostics
8. frontend-design — Production-grade UI generation
...
Recommended:
11. security-guidance — File write scanning. Has PreToolUse hook.
...
Total: X new tools (Y already installed, skipped)
Global: N plugins Local: M plugins
For any plugin with continuous hooks (PreToolUse, PostToolUse, UserPromptSubmit), append a warning after its description:
⚡ Has N continuous hooks — install per-project only
Then ask:
How would you like to proceed?
- Install all Essential tools now
- Install all Essential + Recommended tools now
- Install everything
- Walk me through one by one (ask for each)
- Just show me the list, I'll install manually
Based on the user's choice:
Install commands by scope:
/plugin install <name>@marketplace (default scope — goes to ~/.claude/settings.json)/plugin install <name>@marketplace from within the project directory. Note: Claude Code installs to the project scope when run from a project context.For LSP plugins, also note if the language server binary needs to be installed separately (e.g., npm install -g typescript-language-server).
For hook-heavy plugins, always print the warning:
Note: This plugin has continuous hooks. It is installed locally for this project only.
Re-read ~/.claude/settings.json to get the updated enabledPlugins list. Compare against the recommendations to confirm what was actually installed.
Print the summary:
=== Installation Summary ===
Installed: X new plugins
Already had: Y plugins
Skipped: Z plugins
Your updated toolkit covers:
- Languages: TypeScript, Swift, ...
- Workflow: TDD, code review, debugging, ...
- Security: SAST scanning, vulnerability detection
- Frontend: design, accessibility, browser testing
If any recommended plugin doesn't appear in the updated settings, note it:
Note: plugin-name may not have installed — verify with /plugin list
> /install-recommended-tools
install-recommended-tools v2.0.0
=== Installed Tools ===
Plugins: 12 global, 0 local
MCP servers: 1 configured (playwright)
What types of development do you work on?
> 1, 2
What languages do you primarily use?
> 1, 3
=== Recommended Tools ===
GLOBAL (install once, available everywhere)
Essential:
1. superpowers — TDD, debugging, brainstorming [already installed]
2. code-review — Multi-agent PR code review [already installed]
...
LOCAL (install in this project only)
Essential:
5. typescript-lsp — TS/JS type checking and diagnostics
/plugin install typescript-lsp@claude-plugins-official
6. frontend-design — Production-grade UI generation
/plugin install frontend-design@claude-plugins-official
...
Recommended:
9. security-guidance — File write vulnerability scanning
/plugin install security-guidance@claude-plugins-official
⚡ Has 1 continuous hook — install per-project only
Total: 6 new tools (8 already installed, skipped)
Global: 0 new Local: 6 new
How would you like to proceed?
> 1
/plugin install commands themselves — Claude Code cannot programmatically install plugins.research/developer-tools/claude/plugins-and-skills-catalog.md, relative to repo root)..mcp.json) can be done by Claude Code directly if the user approves.testing
Validate cookbook integrity — frontmatter, cross-references, indexes, skills, rules, file placement. Run from cookbook or consuming project.
tools
Remove the agentic cookbook from your project. Removes rules, state files, CLAUDE.md section, and optionally global plugins and status line.
tools
Analyze a macOS SwiftUI app and plan its conversion to native AppKit. Triggers on 'port to AppKit', 'convert SwiftUI to AppKit', 'remove SwiftUI', or /port-swiftui-to-appkit.
tools
--- name: plan-cookbook-recipe version: 2.3.0 description: Interactively design a new cookbook recipe through guided discussion disable-model-invocation: true context: fork allowed-tools: Read, Glob, Grep, Agent, Write, Edit, AskUserQuestion, Bash(git *), Bash(mkdir *), Bash(ls *) argument-hint: [recipe-name] [--version] --- # Plan Agentic Cookbook Recipe v2.3.0 ## Startup **First action**: If `$ARGUMENTS` is `--version`, print `plan-cookbook-recipe v2.3.0` and stop. Otherwise, print `plan-c