bazzite-ai-dev

Claude Code plugin for Bazzite AI development with enforcement agents and development tools.

Purpose¶

This plugin provides:

1. Development skills for building, testing, and maintaining Bazzite AI
2. Enforcement agents that ensure code quality and policy compliance
3. GitHub MCP integration for repository operations

MCP Server¶

This plugin includes a GitHub MCP server for repository operations.

Tools available:

• Issues: issue_read, issue_write, add_issue_comment, list_issues, search_issues
• Pull requests: pull_request_read, list_pull_requests, search_pull_requests
• Workflows: list_workflows, list_workflow_runs, get_workflow_run, get_job_logs
• Repository: get_file_contents, list_commits, get_commit, list_branches
• Labels: get_label, list_label, label_write

Prerequisites:

• github-mcp-server available via direnv (installed in project)
• GITHUB_TOKEN environment variable set

Available Skills (6)¶

Enforcement Agents¶

These agents are automatically invoked to enforce development policies:

Blocking Agents (Must Pass)¶

Advisory Agents¶

Usage Examples¶

# Build the OS image
/bazzite-ai-dev:build
# Claude will help with image building, troubleshooting, etc.

# Enable development overlay mode
/bazzite-ai-dev:overlay
# Claude will guide you through overlay setup for live justfile editing

# Run runtime verification tests
/bazzite-ai-dev:test
# Claude will help verify GPU, services, and pod functionality

# Clean up after development
/bazzite-ai-dev:clean
# Claude will help clean build artifacts and caches

# Manage Git LFS files
/bazzite-ai-dev:lfs
# Claude will help with large file checkout and verification

# Generate documentation recordings
/bazzite-ai-dev:record
# Claude will help batch-record ujust commands for docs

Installation¶

Manual Loading¶

# Load both plugins for full development experience
claude --plugin-dir ./plugins/bazzite-ai --plugin-dir ./plugins/bazzite-ai-dev

Permanent Configuration¶

Add to your Claude Code settings:

{
  "plugins": [
    "/path/to/bazzite-ai-testing/plugins/bazzite-ai",
    "/path/to/bazzite-ai-testing/plugins/bazzite-ai-dev"
  ]
}

Development Workflow¶

1. Enable overlay testing: just overlay refresh (auto-enables if needed)
2. Make changes to justfiles in just/ directory
3. Refresh overlay: just overlay refresh
4. Test with ujust: ujust <your-command>
5. Verify LOCAL: Check systemctl status, journalctl logs
6. Run pre-commit: pre-commit run --all-files
7. Commit (enforcement agents will verify)

Policies Enforced¶

• LOCAL system verification required before claiming "working"
• ~/.config files are outputs - edit source code instead
• 100% pre-commit hook pass rate required
• All commands must support non-interactive execution
• .just files must be under 30K (split proactively at 25K)
• Never use sudo ujust - handle sudo internally
• Never use just -f for testing - use overlay method
• Never edit pixi.lock manually - regenerate via pixi install

Related¶

• bazzite-ai: OS user skills (separate plugin)
• CLAUDE.md: Full policy documentation
• AGENTS.md: Operational commands and architecture

MCP Server¶

Github (stdio)

• Command: github-mcp-server stdio

Available Tools (23):

Skills¶

Agents¶

Blocking Agents¶

These agents must pass before proceeding with the action.

Advisory Agents¶

These agents provide guidance but do not block actions.