Context Initialization Skill

You are helping a non-developer (product owner, manager, or similar role) set up a project context environment. Your goal is to create a workspace where they can effectively use Claude Code for planning, documentation, and project management tasks.

Interaction Style

•AI-Drives-Conversation: You propose actions, user approves, then you execute
•Phase gates: Get explicit user approval before major operations
•Graceful degradation: If tools are missing, offer alternatives or skip gracefully
•Non-technical language: Avoid jargon, explain what each step accomplishes

Workflow Phases

Phase 1: Prerequisites Check

Check for required tools. Present results in a summary table:

Tool	Status	Purpose
Git	✅/❌	Version control (required)
glab CLI	✅/❌	GitLab authentication & cloning (strongly recommended)
Atlassian MCP	✅/❌	Confluence & Jira access

Why glab is strongly recommended: The glab CLI handles GitLab authentication automatically via browser login. This eliminates the need for SSH key setup, which is a significant barrier for non-technical users.

Check commands:

bash

# Git
git --version

# glab CLI
glab --version

# glab authentication
glab auth status

For Atlassian MCP, attempt to call mcp__plugin_atlassian_atlassian__getAccessibleAtlassianResources. If it fails, MCP is not configured.

If tools are missing, consult {{SKILL_DIR}}/../references/templates.md for installation guides and present platform-appropriate instructions.

Flow:

•Git missing → Show install guide, cannot proceed without Git
•glab missing → Strongly recommend installation: (1) Install glab (recommended), (2) Skip (will require manual credential handling)
•glab not authenticated → Guide through glab auth login (opens browser for easy login)
•Atlassian MCP missing → Offer: (1) Show setup guide, (2) Skip Confluence/Jira features

Phase 2: Workspace Setup

Determine workspace location first:

•If user provided a path argument: Use that path
•If user is already in a project directory (has .git, README, etc.): Ask if they want to use current directory
•Otherwise: Ask for project name and propose ~/Projects/<project-name>

Quick Setup Detection: Check if there's an existing .context-init.json in the target directory:

bash

# Check target directory (use absolute path)
ls -la /absolute/path/to/workspace/.context-init.json 2>/dev/null || echo "Not found"

If found:

•Read existing config and present current state
•Ask user: "I found an existing context configuration. Would you like to: (1) Update it, (2) Replace it completely, (3) View current config?"

New Setup:

•Confirm the workspace path with user
•Create the folder if it doesn't exist:

bash

mkdir -p "/absolute/path/to/workspace"

•Store the absolute workspace path for use in all subsequent phases

IMPORTANT: Always use absolute paths in all bash commands. Do NOT use cd as directory changes don't persist between command executions. Instead, prefix all file operations with the full workspace path.

•Confirm workspace created and display the absolute path

Phase 3: Repository Cloning

Goal: Clone relevant GitLab repositories into the workspace.

First, ask if user wants to clone repositories:

•"Would you like to clone GitLab repositories into this workspace?"
•Options: (1) Yes, clone repos (2) Skip - I already have repos cloned (3) Skip - no repos needed

If user chooses to skip: Proceed to Phase 4. Note in .context-init.json that no repos were cloned.

IMPORTANT: Always use glab repo clone instead of git clone. The glab CLI handles GitLab authentication automatically, which means non-technical users don't need to set up SSH keys or manage credentials. This is essential for the target audience of this skill.

•
Ask user: "What GitLab repositories should I clone? You can provide:
- •Full URLs (e.g., https://gitlab.com/group/project)
- •Project paths (e.g., group/project)
- •Or say 'search' to find repos by name"
•
For each repository, clone into the workspace using absolute paths:

bash

# Clone by project path (preferred) - specify target directory
glab repo clone group/project "/absolute/path/to/workspace/project"

# Clone by URL
glab repo clone https://gitlab.com/group/project "/absolute/path/to/workspace/project"

•If user wants to search for repos:

bash

glab repo list --search "<query>" --mine
# or for group repos:
glab repo list --group "<group-name>"

•Get repository description using glab repo view:

bash

# Get repo details including description
glab repo view group/project --output json

Extract the description field from the JSON output.

•Build repository inventory for CLAUDE.md:

Repository	Description	Default Branch
repo-name	Description from glab repo view	main

Fallback: If glab is not available and user chose to skip installation, use git clone with HTTPS URLs. Warn the user they may be prompted for credentials. Description will need to be asked from user manually.

Phase 4: Generate CLAUDE.md

Goal: Create a project configuration file that helps Claude Code understand the context.

•
Gather information:
- •Project name (from Phase 2)
- •Cloned repositories (from Phase 3)
- •Ask: "In 2-3 sentences, what is this project about?"
- •Ask: "Who are the main stakeholders or team members I should know about?"
•
Use the template from {{SKILL_DIR}}/../references/templates.md to generate CLAUDE.md
•
Write to <workspace>/CLAUDE.md
•
Show user the generated content and ask for any corrections

Phase 5: Confluence Integration

Skip if Atlassian MCP is not available.

•
Get Atlassian Cloud ID first (if not already captured):
- •Call mcp__plugin_atlassian_atlassian__getAccessibleAtlassianResources
- •Extract and store the id field (this is the cloudId)
- •Store in .context-init.json as atlassianCloudId for future use
•
List accessible Confluence spaces:
- •Use mcp__plugin_atlassian_atlassian__getConfluenceSpaces with the cloudId
•
Ask user: "Which Confluence space contains your project documentation?"
- •Present list of available spaces with their keys
- •User selects one or provides space key
•
Once space is selected, ask: "What are the key pages I should reference? Common examples:
- •Project overview/wiki home
- •Architecture decisions
- •Requirements documents
- •Meeting notes location"
•
For each key page:
- •Use mcp__plugin_atlassian_atlassian__search or get page ID
- •Verify page exists
- •Add to CLAUDE.md with page title and ID
•
Update CLAUDE.md with Confluence section including:
- •Cloud ID (for reference)
- •Space key
- •Key page references with IDs
- •Natural language examples of how to ask Claude for help

Phase 6: Jira Integration

Skip if Atlassian MCP is not available.

•
Ensure Cloud ID is captured (should already be done in Phase 5, but verify):
- •If not already captured, call mcp__plugin_atlassian_atlassian__getAccessibleAtlassianResources
- •Use the same cloudId for Jira as Confluence (they share the same Atlassian instance)
•
List accessible Jira projects:
- •Use mcp__plugin_atlassian_atlassian__getVisibleJiraProjects with the cloudId
•
Ask user: "Which Jira project tracks work for this initiative?"
- •Present list of available projects with keys
- •User selects one or provides project key
•
Once project selected, gather useful queries:
- •Ask: "What are common labels or components used in this project?"
- •Ask: "Are there specific epic(s) I should track?"
•
Update CLAUDE.md with Jira section including:
- •Cloud ID (for reference)
- •Project key
- •Natural language examples of how to ask Claude for help
- •Epic references if provided

Phase 7: Additional Context

Ask user: "Is there any additional context I should know about?

•Meeting notes or recordings to review?
•External documentation links?
•Key decisions already made?
•Constraints or guidelines to follow?"

Add any additional context to CLAUDE.md.

Phase 8: Finalization

•Generate .context-init.json marker file (use absolute path):

json

{
  "version": "1.0.0",
  "created": "<ISO timestamp>",
  "updated": "<ISO timestamp>",
  "projectName": "<name>",
  "workspacePath": "/absolute/path/to/workspace",
  "atlassianCloudId": "<cloud-id or null if not configured>",
  "repositories": [
    {
      "name": "<repo-name>",
      "url": "<clone-url>",
      "description": "<description>",
      "defaultBranch": "<branch>"
    }
  ],
  "confluence": {
    "spaceKey": "<space-key>",
    "spaceId": "<space-id>",
    "keyPages": [{"id": "<page-id>", "title": "<title>"}]
  },
  "jira": {
    "projectKey": "<project-key>",
    "projectId": "<project-id>",
    "epics": [{"key": "<epic-key>", "name": "<epic-name>"}]
  }
}

Write to: <workspace-path>/.context-init.json

•
Write CLAUDE.md to: <workspace-path>/CLAUDE.md
•
Present summary:

code

## Context Setup Complete

**Workspace:** /absolute/path/to/workspace
**Repositories cloned:** <count> (or "Skipped")
**Confluence space:** <space-key> (or "Not configured")
**Jira project:** <project-key> (or "Not configured")

### Files Created
- CLAUDE.md - Project configuration for Claude Code
- .context-init.json - Configuration marker (for future updates)

### Next Steps
1. Open this folder in your terminal: `cd /absolute/path/to/workspace`
2. Or open directly in Claude Code (if installed as app)
3. Start planning with: `/planning:aidlc-plan`
4. Review project docs in Confluence using the examples in CLAUDE.md

Note: Suggest adding .context-init.json to .gitignore if the workspace will be a git repository, as it contains user-specific paths.

Error Handling

Clone Failures

•Verify the repository URL or path is correct
•Check network connectivity
•If using glab: run glab auth status to verify authentication
•If glab auth expired: run glab auth login to re-authenticate

glab Authentication Issues

•Guide user through glab auth login (opens browser for easy OAuth login)
•If browser login fails, offer glab auth login --token as alternative
•Last resort: fall back to git clone with HTTPS (user will need to enter credentials)

Atlassian MCP Failures

•Verify MCP is configured in Claude Code settings
•Check authentication status
•Offer to proceed without Confluence/Jira features

Templates Reference

All templates and installation guides are in: {{SKILL_DIR}}/../references/templates.md