Automated setup for Cloud Security Alliance development environments on macOS and Windows. Get from a bare machine to running AI coding assistants in a few commands.
Installs Claude Desktop, ChatGPT Desktop, Claude Code, Codex CLI, Gemini CLI, plus 1Password (GUI + CLI) for secret management in AI workflows. The install scripts walk you through GitHub login (gh auth login) and configure your Git identity from your GitHub profile. They also detect tools installed via the wrong method (e.g., Claude Code via Homebrew or npm) and migrate them to the correct installer.
bash -c "$(curl -fsSL -H 'Cache-Control: no-cache' https://raw.githubusercontent.com/CloudSecurityAlliance/DesktopSetup/HEAD/scripts/macos-ai-tools.sh)"The macOS script shares a base layer (Xcode CLI Tools, Homebrew, Node.js/npm, Python) and installs it if not already present. It also installs the document toolchain — pandoc and typst, plus pyyaml and pymupdf in ~/.default_venv — which the document-pipeline plugin needs to render CSA PDFs.
Note: The
-H 'Cache-Control: no-cache'flag forces a fresh download from GitHub — without it, a stale copy from the CDN edge cache can persist for a few minutes after we ship fixes.
Windows requires one-time PowerShell setup before running any script in this repo.
Not your personal machine? Changing the execution policy is a security setting. If this is a work laptop managed by your IT department, ask them for permission before proceeding.
Step 1 — Check your current policy. Open PowerShell as Administrator (press the Windows key, type powershell, right-click Windows PowerShell, and select Run as administrator). Then run:
Get-ExecutionPolicyNote the value it returns (usually Restricted on a fresh install). You'll restore this afterwards.
Step 2 — Temporarily allow script execution:
Set-ExecutionPolicy RemoteSignedWhen prompted "Do you want to change the execution policy?", type Y and press Enter.
Step 3 — Run the AI tools script. Close the Administrator window and open a regular PowerShell window:
irm https://raw.githubusercontent.com/CloudSecurityAlliance/DesktopSetup/HEAD/scripts/windows-ai-tools.ps1 -Headers @{'Cache-Control'='no-cache'} | iexThe Windows AI tools script also installs Git, GitHub CLI, Python, Node.js, pandoc, and typst (via winget) alongside the AI apps and CLIs, plus pyyaml and pymupdf via pip for the document-pipeline plugin. Requires Windows 10/11 and winget.
Note: The
-Headers @{'Cache-Control'='no-cache'}flag forces a fresh download from GitHub — without it, a stale copy from the CDN edge cache can persist for a few minutes after we ship fixes.
Step 4 — Restore the original policy. Once you're done running scripts, re-open PowerShell as Administrator and set the policy back to whatever Step 1 reported:
Set-ExecutionPolicy RestrictedReplace Restricted with the value from Step 1. If you plan to run PowerShell scripts regularly, you can leave it as RemoteSigned.
Once AI tools are installed, use these one-liners to clone any CSA repo and launch Claude Code. Replace ORG/REPO with the actual org and repo name.
bash -c "$(curl -fsSL -H 'Cache-Control: no-cache' https://raw.githubusercontent.com/CloudSecurityAlliance/DesktopSetup/HEAD/scripts/clone-and-claude.sh)" -- ORG/REPO$env:CSA_REPO='ORG/REPO'; irm https://raw.githubusercontent.com/CloudSecurityAlliance/DesktopSetup/HEAD/scripts/clone-and-claude.ps1 -Headers @{'Cache-Control'='no-cache'} | iexThe scripts check prerequisites, clone the repo to ~/GitHub/OrgName/RepoName, and tell you how to launch Claude Code. Safe to re-run — they skip the clone if the repo already exists and pull latest changes instead.
Run the update script to update everything at once (Homebrew formulas/casks, npm globals, pip packages, and Claude Code). Saves a snapshot of all installed versions before updating so you can roll back if anything breaks:
bash -c "$(curl -fsSL -H 'Cache-Control: no-cache' https://raw.githubusercontent.com/CloudSecurityAlliance/DesktopSetup/HEAD/scripts/macos-update.sh)"Plugins only — if you just want to refresh CSA Claude Code plugins (no Homebrew/npm/pip), run the standalone plugins script:
bash -c "$(curl -fsSL -H 'Cache-Control: no-cache' https://raw.githubusercontent.com/CloudSecurityAlliance/DesktopSetup/HEAD/scripts/macos-plugins.sh)"Or update individual package managers manually:
brew update && brew upgrade # Homebrew formulas and casks
npm update -g # npm global packages (Codex, Gemini, Wrangler)
pip install --upgrade pip # pip itself
claude update # Claude CodeSnapshots are saved to ~/Library/Logs/CSA-DesktopSetup/ with timestamps.
Re-run either install script to upgrade — they detect what's already installed and update in place. npm tools can also be updated manually:
npm update -g @openai/codex @google/gemini-cliClaude Code updates itself automatically.
Plugins only — Windows counterpart of macos-plugins.sh:
irm https://raw.githubusercontent.com/CloudSecurityAlliance/DesktopSetup/HEAD/scripts/windows-plugins.ps1 -Headers @{'Cache-Control'='no-cache'} | iexOptional — install these alongside AI tools for a complete work setup. Covers productivity, communication, browser, and (optionally) general developer tools.
Installs: 1Password, Slack, Zoom, Chrome, Microsoft Office, Git, GitHub CLI. Optional dev profile adds VS Code, AWS CLI, and Wrangler.
bash -c "$(curl -fsSL -H 'Cache-Control: no-cache' https://raw.githubusercontent.com/CloudSecurityAlliance/DesktopSetup/HEAD/scripts/macos-work-tools.sh)"Installs: Git, GitHub CLI, 1Password, Slack, Zoom, Chrome. Optional dev profile adds VS Code, AWS CLI, and Wrangler. Requires the PowerShell execution policy from the Windows AI tools section above.
irm https://raw.githubusercontent.com/CloudSecurityAlliance/DesktopSetup/HEAD/scripts/windows-work-tools.ps1 -Headers @{'Cache-Control'='no-cache'} | iexThe AI tools installers (and the macOS updater) install a curated set of Claude Code plugins so CSA staff can use them right away and explore what's possible. Public plugins install for everyone; CSA-marketplace plugins install only if your GitHub account can access the private CSA marketplaces.
- superpowers — brainstorming, writing-plans, TDD, systematic-debugging, code review, verification, dispatching-parallel-agents, using-git-worktrees
- feature-dev —
/brainstorm,/write-plan,/execute-plan,/implement,/finish-branch - claude-code-setup — claude-automation-recommender (scan a repo, recommend hooks/agents/skills)
- claude-md-management —
/initimprovements, claude-md-improver - session-report — HTML report of tokens, cache, skills, expensive prompts
- explanatory-output-style — toggle: add educational explanations to responses
- learning-output-style — toggle: interactive learning with contribution requests
- commit-commands —
/commit,/commit-push-pr,/clean_gone - code-review —
/review - pr-review-toolkit —
/review-pr, multi-agent PR review (silent-failure-hunter, type-design-analyzer, pr-test-analyzer, comment-analyzer, code-simplifier) - github — GitHub MCP (issues, PRs, code search, releases, reviews)
- frontend-design — polished UI code that avoids generic AI aesthetics
- playwright — browser automation and UI testing via Playwright MCP
- chrome-devtools-mcp — Chrome DevTools MCP (debugging, performance, a11y, LCP)
- playground — single-file interactive HTML explorers
- typescript-lsp — TypeScript language server integration
- security-guidance — security review and guidance
- cloudflare — Workers, Pages, D1/R2/KV, Durable Objects, Agents SDK, Wrangler
- claude-api — Claude API / Anthropic SDK (migrations, caching, tool use, batching)
- agent-sdk-dev —
/new-sdk-app, Agent SDK verifiers - mcp-server-dev — build-mcp-server, build-mcp-app, build-mcpb (local .mcpb bundles)
- plugin-dev —
/create-plugin, plugin-structure, command/agent/skill/hook development - skill-creator — create, edit, and benchmark skills
- pydantic-ai — Pydantic AI framework
- slack —
/standup,/find-discussions,/summarize-channel,/draft-announcement,/channel-digest - document-skills — docx, pptx, pdf, xlsx, canvas-design, brand-guidelines, internal-comms, theme-factory, webapp-testing
- example-skills — reference implementations of the document skills above
- cwe-analysis — CWE assignment, chains, AI relevance
- incident-analysis — OSINT, timeline, impact, defensive recs for cloud/AI incidents
- nist-ir-8477-mapping — map between frameworks using NIST IR 8477
- security-knowledge-ingestion — convert standards/regs into structured data
- document-pipeline — Markdown to a branded, accessible CSA PDF (render, preflight, copy edit, design review). Needs
pandocandtypst— see its README - vendor-research — guided vendor evaluation and procurement decisions
- csa-spreadsheets — parse CCM/AICM spreadsheets into JSON, CSV, or Markdown
- cino-project-tracker — CINO Airtable project registry
- audience-lens — build audience profiles for writing tasks
- writing-style-forge — generate writing-style plugins from samples
- research-initiative-tracker — CSA research initiative tracking
- csa-certification-development, csa-training-content-development, csa-training-design-system — training & certification workflows
The plugin lists live in scripts/csa-plugins.txt
and scripts/csa-plugins-internal.txt
— edit those to change what gets installed by default. Users can
disable any individual plugin locally with claude plugin disable <name>.
Re-run the same command with CSA_DEBUG set. Everything still prints to the screen, and a full
transcript — every command, its output, and its exit code — is written to a file in your home
directory. It works on every script here, not just the two shown.
CSA_DEBUG=1 bash -c "$(curl -fsSL -H 'Cache-Control: no-cache' https://raw.githubusercontent.com/CloudSecurityAlliance/DesktopSetup/HEAD/scripts/macos-ai-tools.sh)"The CSA_DEBUG=1 goes before bash, as a prefix on the same line. That is what makes it
reach the script.
Paste these one at a time. Copying both lines together and pasting them into the console has been observed to close the window immediately, before either line runs. Two separate pastes, or the single-line form further down, both avoid it.
First, turn logging on:
$env:CSA_DEBUG = '1'Then run the script:
irm https://raw.githubusercontent.com/CloudSecurityAlliance/DesktopSetup/HEAD/scripts/windows-ai-tools.ps1 -Headers @{'Cache-Control'='no-cache'} | iexThe variable stays set for the rest of that PowerShell window, so a second run also logs. To stop logging again:
Remove-Item Env:\CSA_DEBUGOr as a single line, if you would rather paste once — same thing, joined with ;, which is
the form this README already uses for CSA_REPO below:
$env:CSA_DEBUG = '1'; irm https://raw.githubusercontent.com/CloudSecurityAlliance/DesktopSetup/HEAD/scripts/windows-ai-tools.ps1 -Headers @{'Cache-Control'='no-cache'} | iexThere is no -Debug switch, and the two obvious guesses both go wrong:
| what you might type | what actually happens |
|---|---|
irm ... --Debug ... | iex |
fails immediately — "a positional parameter cannot be found that accepts argument '--Debug'" |
irm ... -Debug ... | iex |
runs, and logs nothing. -Debug is a real parameter on irm: it applies to the download, not to the script iex then executes |
The reason there is no switch: irm … | iex fetches text and executes it, so the script never
receives an argument vector for a flag to arrive in. An environment variable is the only thing
that crosses that boundary — which is also how NONINTERACTIVE works here.
Either spelling is accepted, since $env: is easy to forget — these are alternatives, not a
sequence, so use one or the other:
$env:CSA_DEBUG = '1'$CSA_DEBUG = '1'So is any of 1, true, yes or on, in any case.
The path is printed at the start and again at the end:
~/desktopsetup-YYYYMMDD-HHMMSS.log (macOS)
C:\Users\<you>\desktopsetup-YYYYMMDD-HHMMSS.log (Windows)
Readable only by you (mode 0600 / an ACL granting just your account). One file covers the whole run, including the CSA-internal setup that runs as a separate process — so there is only ever one file to send.
It exists even if the run stops early. A script that refuses to proceed (wrong platform, running as Administrator, a missing prerequisite) still leaves a log saying which check refused, because that is exactly when you want one.
Known credential shapes — tokens, client secrets, bearer headers, refresh tokens — are replaced
with <redacted> before anything is written, and the CSA OAuth client is never captured at all.
That is a safety net, not a guarantee: read the file before you send it to anyone. The first
line of every log says the same.
macos-mcp-setup.sh discovered Airtable and GitHub personal access tokens and wrote them
into the Claude Code, Codex, and Gemini configs. It is retired — those services are now
reachable as hosted connectors over OAuth, with no long-lived token stored on disk. The
script is kept in archives/ for reference.
If you ran it, you may still have tokens on disk. Check for airtable or github MCP
entries carrying a bearer token in ~/.claude.json, ~/.codex/config.toml, and
~/.gemini/settings.json, and remove any you no longer use — then revoke the token at the
service. Nothing removes them for you.
Each script is self-contained and idempotent (safe to re-run):
macos-ai-tools.sh— AI desktop apps and coding assistants with migration support (macOS)macos-update.sh— Update all installed tools with version snapshots (macOS)macos-plugins.sh— Standalone Claude Code plugin install/update (macOS)windows-plugins.ps1— Standalone Claude Code plugin install/update (Windows)macos-work-tools.sh— Core work apps + optional developer tools (macOS)windows-ai-tools.ps1— AI desktop apps and coding assistants with migration support (Windows)windows-work-tools.ps1— Core work apps + optional developer tools (Windows)clone-and-claude.sh— Clone a CSA repo and set up for Claude Code (macOS)clone-and-claude.ps1— Clone a CSA repo and set up for Claude Code (Windows)
Design documents and implementation notes for the scripts.
Previous versions of scripts preserved for reference.
Found a problem? Have a suggestion?
Open an issue — we have templates for common requests.
Apache License 2.0 — see LICENSE.