codingthefuturewithai/teamcraft-jcg:teamcraft-setup

Goal

Get Teamcraft JCG fully operational. That means the Atlassian MCP server working (Jira + Confluence), GitHub CLI authenticated (for developers), and the right companion plugins installed. This skill verifies each connection, guides through setup if anything is missing, and leaves the user ready to run any other Teamcraft skill.

Run this once when first installing the plugin. Re-run it anytime a connection stops working.

Step 1: Understand the User's Context

Before verifying any connections, understand who is setting up and what they need. Ask:

"Before we start, a couple of quick questions so I can tailor the setup:

1. What kind of work will you be doing with Teamcraft? (e.g., writing code, planning projects, managing requirements, a mix of everything)
2. Are you running this in Claude Code or Claude Cowork?"

What this tells you:

• What they do determines which tools matter. If they write and ship code, they'll need GitHub CLI and LSP plugins. If they focus on planning, requirements, or documentation, they only need Atlassian MCP and companion plugins — upstream skills don't require the build loop tools.
• Environment determines how to guide MCP setup if something is missing. In Claude Code, the claude mcp add CLI works. In Claude Cowork, MCP servers must be configured in Claude Desktop's config file — follow references/cowork-mcp-setup.md.

Store what you learn for the rest of the setup flow. If the user's answers make it obvious (e.g., "I'm planning sprints in Cowork"), don't ask redundant follow-ups.

Step 2: Verify Atlassian MCP (Jira + Confluence)

Test Jira connectivity by calling mcp__sooperset-mcp-atlassian__jira_get_all_projects. If that succeeds, Jira is working.

Test Confluence connectivity by calling mcp__sooperset-mcp-atlassian__confluence_search with a broad query. If that succeeds, Confluence is working.

If either fails, go to references/atlassian-mcp-setup.md for guided setup.

Step 3: Verify GitHub CLI

Skip this step if the user doesn't write or ship code. GitHub CLI is only needed for the build loop (fetch-issue, plan-and-implement-issue, complete-issue). Upstream skills like capture-requirements, plan-sprint, and onboard don't use it.

If the user does write code:

1. If in Claude Code: run gh auth status via Bash. If it succeeds, GitHub CLI is authenticated. If gh is not found, guide the user: install from https://cli.github.com/ then run gh auth login.
2. If in Cowork: note that GitHub CLI verification requires Claude Code. Build loop skills need it; upstream skills do not. The user will need to set this up in a Claude Code session.

Step 4: Check Companion Plugins

Based on the environment from Step 1:

Claude Code

Run claude plugin list and check for the following. If missing, tell the user what's needed and why. Do not assume specific install commands are correct — plugin registries and install mechanisms evolve. Name the plugin and let Claude Code guide the installation.

context7 — plan-and-implement-issue uses it for real-time library documentation during technical research. Without it, tech research falls back to training data. Note: context7 may be available as a plugin or as an MCP server — check both if one doesn't work.

frontend-design — capture-requirements uses it for higher-fidelity client-facing HTML mockups. Without it the session still works but visuals look generic.

LSP plugins — Only if the user writes code. Follow references/lsp-plugins.md.

Cowork

Tell the user which plugins are needed and why. Help them figure out how to install plugins in their Cowork environment.

Step 5: Done

Summarize what's working and what (if anything) was skipped or still needs attention. Then let the user know setup is complete and Teamcraft is ready to use.

If the user asks what to do next, point them to teamcraft-jcg:learn-teamcraft for a walkthrough of available skills. Don't steer them toward a specific activity — that's their call.