8.12.0

Fixed

• coldbox ai doctor and coldbox ai uninstall checking wrong directory
  • Both commands were checking for a .ai directory instead of .agents, causing them to always report "not installed" even after a successful coldbox ai install
  • AIService.diagnose() now uses the static.AI_DIR constant (.agents) instead of the hardcoded /.ai path
  • coldbox ai uninstall now correctly checks, removes, and references the .agents directory
• coldbox ai skills add slug --list was not working.
• coldbox ai skills remove reinstalling removed skills during refresh
  • When a skill was removed via coldbox ai skills remove, the subsequent agent config regeneration step (refresh()) would detect the skill as "missing" (because its module dependency was still present in box.json) and immediately reinstall it
  • Removed skills are now tracked in a new manifest.excludes[] array so that refresh() will never auto-reinstall them
  • Explicitly re-installing a previously excluded skill via coldbox ai skills install lifts the exclusion and restores normal auto-management

Added

• VSCode Copilot MCP Mirroring
  • When Copilot is a configured agent, MCP server configuration is now mirrored to .vscode/mcp.json using the VSCode-specific schema ("servers" + "inputs": [])
  • Ensures GitHub Copilot agents in VSCode can discover MCP servers registered via coldbox ai mcp commands
  • .vscode/mcp.json is written alongside the root .mcp.json whenever generateMCPJson() runs (install, refresh, MCP add/remove)
• coldbox ai skills list --json flag to output the skills manifest in JSON format for easier parsing in scripts and CI pipelines
• coldbox ai skills update command to re-download and overwrite all installed registry skills with per-skill feedback and progress updates
• coldbox ai skills update <name> command to re-download and overwrite a single installed registry skill by local name

Improvements

• New progress bar when doing coldbox ai skills list --outdated to check for updates in the registry, providing better feedback during potentially long-running integrity checks