ProdE Skill

The ProdE skill is a packaged set of instructions that teaches your coding agent how to work with ProdE. It tells the agent which ProdE tools exist, the order to use them in, and when each one is the right call, so you get good results without spelling out the workflow yourself every time.

It ships as a single prode skill with role playbooks inside it. The agent reads the entry point, matches your task to a playbook (asking how something works, support and troubleshooting, data and schema questions, marketing claims, solutions and RFPs, developer work across repos, planning a project into ProdE, or implementing one that is already planned), then loads only that playbook.

What the installer does

On macOS and Linux, one command handles the whole setup:

1. Logs you in to ProdE with a device code. It prints a URL and a code, and you approve in the browser.
2. Installs the skill for each agent you target.
3. Connects the ProdE MCP server for that agent, using the token from step 1.
4. Registers hooks that keep the skill current and pre-approve ProdE tool calls, so you are not asked to confirm every one.

There is no second command to run afterwards, and nothing to configure by hand.

Requirements

• macOS or Linux (Ubuntu is what gets tested). Windows users: see Install on Windows.
• bash, curl, tar, and a working Python 3 on your PATH. On macOS, Python 3 comes with the Xcode Command Line Tools (xcode-select --install); on Ubuntu, sudo apt install python3.
• A ProdE account. The login step is part of the install.

Install on macOS or Linux

curl -fsSL https://curiousbox-public.nyc3.cdn.digitaloceanspaces.com/skills/install-prode.sh | bash

With no flags, the installer targets whichever supported agents it finds already set up on the machine, and falls back to Claude Code when it cannot detect any.

To choose the agents yourself, pass --agent. It takes claude, cursor, codex, a comma-separated list, or all:

Claude Code

curl -fsSL https://curiousbox-public.nyc3.cdn.digitaloceanspaces.com/skills/install-prode.sh | bash -s -- --agent claude

Cursor

curl -fsSL https://curiousbox-public.nyc3.cdn.digitaloceanspaces.com/skills/install-prode.sh | bash -s -- --agent cursor

Both Claude Code and Cursor

curl -fsSL https://curiousbox-public.nyc3.cdn.digitaloceanspaces.com/skills/install-prode.sh | bash -s -- --agent claude,cursor

Every supported agent (Claude Code, Cursor, and Codex)

curl -fsSL https://curiousbox-public.nyc3.cdn.digitaloceanspaces.com/skills/install-prode.sh | bash -s -- --agent all

Restart your agent afterwards so it picks up the new skill and hooks.

Where things land

The skill installs into your home directory, so it is available across all your projects. There is no per-project install.

Codex uses the ~/.agents open standard rather than a directory of its own, and one Codex config serves the CLI, the IDE extension, and the desktop app. Its MCP entry is the one thing left to you: the format differs between Codex versions, so the installer shows you the exact block instead of guessing.

Everything else the installer owns lives in ~/.prode/: your token, your cached repository list, the deployment this machine talks to, and an offline copy of the uninstaller.

The installer also adds a short, clearly fenced ProdE section to your agent's instructions file, telling it to reach for ProdE on codebase questions. For Claude Code that is ~/.claude/CLAUDE.md, for Codex ~/.codex/AGENTS.md. Cursor has no home-level instructions file, so it gets the same block per workspace instead. The rest of the file is left byte for byte as it was, and re-installs replace only the fenced section.

Options

On-prem and self-hosted

If your ProdE runs on your own infrastructure, point the installer at your backend with --api. Give it the origin, and the rest of the path is worked out for you:

curl -fsSL https://curiousbox-public.nyc3.cdn.digitaloceanspaces.com/skills/install-prode.sh | bash -s -- --api https://prode.your-company.com

That routes login and repository lookup to your deployment, and records the choice machine-wide, so later re-runs and background updates stay on it. The skill package itself comes from the shared CDN either way.

Install on the claude.ai web app

The web app takes an uploaded skill package rather than a script.

1. Download the skill: prode.zip
2. In Claude, go to Settings → Capabilities and enable Code execution and file creation.
3. Go to Customize → Skills, choose + → Upload a skill, and pick prode.zip.

Team and Enterprise admins can upload once for the whole organisation under Organization settings → Skills → + Add.

This route installs the skill files only. Connect the ProdE connector separately, and re-upload the zip when you want a newer version.

Install on Windows

On Windows, download the skill package and extract it into your agent's skills directory.

1. Download the skill: prode.zip
2. Extract it so that the prode folder lands in the right place for your agent:
   • Claude Code: %USERPROFILE%\.claude\skills\prode\
   • Cursor: %USERPROFILE%\.cursor\skills\prode\

Or in PowerShell:

$dest = "$HOME\.claude\skills"          # or "$HOME\.cursor\skills"
New-Item -ItemType Directory -Force $dest | Out-Null
curl.exe -fsSL https://curiousbox-public.nyc3.cdn.digitaloceanspaces.com/skills/prode/latest/prode.zip -o "$env:TEMP\prode.zip"
Expand-Archive -Force "$env:TEMP\prode.zip" -DestinationPath $dest

This is the skill files only: no login, no MCP registration, and no auto-update. Configure the MCP server yourself, and re-run the snippet to update.

Updating

Installs made with the script update themselves. A hook checks for a new version when a session starts, at most a few times a day, and installs it in the background; the new version becomes active in your next session. A failed check leaves the working version exactly as it is.

Re-running your original install command also works, and is the quickest way to pick up a release immediately.

For the claude.ai and Windows routes, updating means fetching prode.zip again.

Uninstalling

Two commands, with different scopes.

Remove one agent's install, stay logged in:

~/.prode/bin/prode-login.sh --uninstall --agent cursor   # or --agent all

This removes that agent's skill folder, its hooks, and its MCP registration.

Log out machine-wide:

~/.prode/bin/prode-login.sh --logout

This removes your stored token and cached repository list, and unregisters the hooks and the MCP server for every agent. It asks whether to remove the skill folders too and keeps them by default. To answer up front, which is useful in scripts, add --remove-skills none, --remove-skills all, or a comma list of agents.

For the claude.ai route, remove the skill under Customize → Skills. For a manual or Windows install, delete the skill folder.