Init Unblocked — One-Time Setup

This skill runs once to set up Unblocked on your machine: installs the CLI, authenticates your account, and configures the MCP server for your AI coding agent.

IMPORTANT — Agent Behavior Contract: Before executing ANY action in this skill (shell commands, opening browsers, modifying config), you MUST:

Tell the user exactly what you are about to do and why

Wait for the user to acknowledge or approve before proceeding

Show the result and confirm success before moving to the next step

Never silently execute commands. Every step should feel like a guided conversation.

Step 0: Pre-Flight Check

Before starting, verify the environment:

Tell the user: "Before we begin, I'm going to check whether the Unblocked CLI is already installed on your machine, and confirm that prerequisites (curl, bash) are available."

Run these checks and report the results:

Unblocked CLI: which unblocked 2>/dev/null && unblocked --version || echo "Unblocked CLI not found"
Prerequisites: which curl && which bash

If the Unblocked CLI is already installed:

Tell the user: "It looks like the Unblocked CLI is already installed (version X). Let me check if it's also authenticated and the MCP server is configured."
Run unblocked help to discover the auth status / whoami command
Check authentication status
If both CLI and auth are in place, skip directly to Step 4: Verify Setup
If CLI is installed but not authenticated, skip to Step 3: Authenticate

If the Unblocked CLI is NOT installed:

Tell the user: "The Unblocked CLI isn't installed yet. I'll walk you through the installation next."
Proceed to Step 1: Install the Unblocked CLI

Step 1: Install the Unblocked CLI

Tell the user:

"I'm going to install the Unblocked CLI by running a shell script from dev.getunblocked.com. This will download and configure the unblocked command-line tool on your machine. Here's the exact command I'll run:"
curl -fsSL https://getunblocked.com/install.sh | bash

Wait for user acknowledgement before running the command.

After the command completes:

Show the user the full output
Confirm whether the install succeeded or failed
If it failed, troubleshoot based on the error output before continuing

Step 2: Discover Available Commands

Tell the user:

"Now I'm going to run unblocked help to see all the commands available in the CLI. This will help me understand what setup steps are needed."

Run: unblocked help

Review the output and identify:

The authentication/login command
The MCP configuration command
Any other setup-related commands

Share a brief summary with the user:

"Here are the key commands we'll use next: [list the relevant commands you found]"

Step 3: Authenticate with Unblocked

Tell the user:

"I'm going to open a browser session now to login with Unblocked for Authorization. This will open your default browser to the Unblocked login page where you can sign in or create an account. The CLI will wait for the browser authentication to complete."

Wait for user acknowledgement before opening the browser.

Run the auth command with the --skip-prompt flag so the browser opens immediately without waiting for interactive input (since we're running from an agent session). For example:

unblocked auth --skip-prompt

Use whichever auth command you discovered in Step 2, but always append --skip-prompt.

After running:

Let the user know the browser should have opened
Wait for them to confirm they completed the login
Check the command output for success/failure
If authentication failed, help troubleshoot

Step 3b: Configure MCP Server

Based on the commands discovered in Step 2, determine if there is an explicit MCP setup/configuration step.

Tell the user:

"Now I'm going to configure the Unblocked MCP server so your AI coding agent can access your team's context. Here's what I'll run: [exact command]"

Wait for user acknowledgement before running.

After running:

Show the output
Confirm the MCP server is configured

Step 4: Verify Setup

Tell the user:

"Let me verify everything is set up correctly before we continue. I'm going to run a few checks."

Run these verification checks (reporting each one to the user):

CLI installed: unblocked --version
Authentication valid: Run the appropriate auth status/whoami command (discovered from unblocked help)
MCP server configured: Check that the MCP tools are accessible by attempting to use one (e.g., unblocked_context_engine or similar)

Report a summary:

"Here's your setup status:"

CLI: [installed / version]

Auth: [authenticated as X / not authenticated]

MCP: [configured / not configured]

If anything failed, help the user fix it before continuing.

If everything passes, tell the user:

You're all set! Unblocked is installed, authenticated, and your MCP server is configured. Every AI coding agent you use now has access to your team's full context — PRs, Slack discussions, Jira tickets, docs, and code history — all queryable in natural language.

⚠️ Important: You need to restart your agent session for the MCP server we just configured to become available. Close this session and start a new one — the Unblocked MCP tools won't be visible to your agent until you do.

Here's what changes now: With Unblocked providing context, you can take on much more ambitious work with your AI agent. Instead of small, isolated tasks, you can tackle full features, complex refactors, and cross-system changes — because the agent understands why your code exists, not just what it does.

Next step: restart your agent session. Once you're in a fresh session, the Unblocked MCP tools will be available and you can start using the /unblock skill by typing commands like:

/unblock implement [feature name]

/unblock refactor [system] to [goal]

/unblock investigate and fix [bug description]

The more context Unblocked has about your codebase, the more ambitious you can be. Ask it anything — after you restart!

What You Can Do After Restarting

Once you restart your agent session, Unblocked will be fully active. Your AI agent will have access to your team's full knowledge graph — PRs, Slack threads, Jira tickets, design docs, and code history. This means you can stop scoping work to what the agent can see in your local files and start thinking bigger.

Use the `/unblock` skill

The /unblock skill is your primary workflow. It hydrates your plan with real context from your team's history and uses Unblocked throughout implementation — from planning to code generation to review.

Example prompts (use these in your new session):

/unblock implement [feature name]
write me a plan for how we can approach [project] with /unblock
/unblock refactor [system] to [goal]
/unblock investigate and fix [bug description]

Think Bigger

Without context, AI agents are limited to what they can see in your repo right now. With Unblocked, they know:

Why code was written the way it was — from PR discussions and Slack threads
What was tried before — so you don't repeat failed approaches
Who the experts are — and what they've said about the system
What the requirements actually are — from Jira, Linear, and design docs

This means you can confidently ask your agent to take on full features, cross-system refactors, and complex debugging — not just small isolated edits.

Resources

Unblocked - Automated context retrieval for engineering teams
Unblocked MCP Docs - Connect Unblocked to your AI coding tools
Unblocked API - Programmatic access to context

init-unblocked

Init Unblocked — One-Time Setup

Step 0: Pre-Flight Check

Step 1: Install the Unblocked CLI

Step 2: Discover Available Commands

Step 3: Authenticate with Unblocked

Step 3b: Configure MCP Server

Step 4: Verify Setup

What You Can Do After Restarting

Use the /unblock skill

Think Bigger

Resources

Use the `/unblock` skill