ColdBox CLI 8.11 is here, and it's one of the most significant releases we've shipped for AI-assisted development. This release wires the CLI directly into our brand new public skills directory at skills.boxlang.io, brings our AI tooling in line with industry-wide agent conventions, and introduces a wave of quality-of-life improvements that make AI integration feel less like setup and more like infrastructure.

If you've been following the BoxLang AI story, you already know we believe AI agents should be first-class citizens in your development workflow — not an afterthought bolted on with copy-pasted prompts. ColdBox CLI 8.11 takes that conviction and ships it. 🎯

📚 Want the full skills directory story? Check out our companion announcement: Introducing skills.boxlang.io — The Public Skills Directory for the BoxLang Ecosystem

🌐 Connected to skills.boxlang.io

The headline feature: ColdBox CLI 8.11 now talks directly to skills.boxlang.io, our new public registry for AI skills across the BoxLang and ColdBox ecosystem.

Skills are on-demand coding cookbooks — focused, task-specific guides that AI agents pull when they need them. Think "how to build a REST API in ColdBox," "how to write a TestBox BDD spec," or "how to wire up cbSecurity passkeys." Instead of dumping the entire framework documentation into your agent's context window, skills are loaded surgically when relevant.

With 8.11, every coldbox ai skills command is now backed by the directory:

# Browse the entire registry interactively
coldbox ai skills install --list

# Pre-filter to a specific publisher or category
coldbox ai skills install --list coldbox/skills
coldbox ai skills install --list coldbox/skills/coldbox-testing

# Search the registry by keyword
coldbox ai skills find "rest api"

This is what we mean when we say the era of skills has come to every ColdBox and BoxLang app. You no longer need to hand-craft prompts or maintain your own internal cookbook — the directory gives you a curated, versioned, searchable library of AI skills that your agents can pull on demand. And module authors can publish their own skills to the directory so their consumers automatically get rich AI context the moment they install the module. 🔥

📋 AGENTS.md: Following the Industry Standard

The AI agent space has converged on a shared convention: a single AGENTS.md file at the project root that any agent — Codex, OpenCode, GitHub Copilot, and others — can read for project context. ColdBox CLI 8.11 fully embraces this standard.

GitHub Copilot has migrated to AGENTS.md following GitHub's August 2025 changelog. This means:

• One source of truth for Copilot, Codex, and OpenCode
• No more .github/copilot-instructions.md divergence
• Cleaner project root, less duplication
• Easier onboarding for new agents that adopt the standard

If you have an existing .github/copilot-instructions.md, the CLI doesn't auto-delete it — run coldbox ai refresh and the new AGENTS.md will be generated. You can clean up the old file at your leisure.

📁 .ai/ → .agents/: Aligning with the Broader AI Ecosystem

The AI tooling world is standardizing around .agents/ as the canonical directory for agent configuration, skills, and guidelines. ColdBox CLI 8.11 makes that move:

• Old: /.ai/
• New: /.agents/

This isn't just cosmetic. It aligns ColdBox projects with the same conventions used across the broader agent tooling ecosystem — from skill repositories to other AI-aware frameworks — so your project structure feels native no matter which agent or tool a developer reaches for.

All internal code paths, CLI commands, and documentation now reference /.agents/. After upgrading, run coldbox ai refresh and the CLI will regenerate everything in the right place.

🔁 Auto-Install Module Skills on Refresh

This is the change that makes skills feel like infrastructure rather than setup.

When you add a module to box.json, coldbox ai refresh now automatically detects and installs the module's corresponding skill from the registry. This mirrors the existing MCP server auto-detection behavior, so skills and servers are always in sync with your project's dependencies.

box install qb
coldbox ai refresh
# ✅ qb guideline added
# ✅ qb skill auto-installed from skills.boxlang.io
# ✅ qb MCP server registered in .mcp.json

Wire this into your box.json and forget about it:

{
  "scripts": {
    "postInstall": "coldbox ai refresh",
    "postUpdate": "coldbox ai refresh"
  }
}

Now every time you install a module, your AI context catches up automatically. No manual copy-paste, no stale documentation, no drift. 💚

🩹 Auto-Recovery of Missing Skills

Previously, if a core skill failed to reinstall during a refresh, it would be silently removed from the manifest. Permanently. That was a footgun, and it's gone.

ColdBox CLI 8.11 now detects and reinstalls any missing core skills (boxlang, coldbox, testbox, commandbox) on every refresh. If something disappears — whether from an interrupted install, a manual deletion, or a manifest hiccup — the next coldbox ai refresh heals it automatically.

coldbox ai refresh
# 🩹 Detected missing skill: boxlang
# ✅ Reinstalled boxlang from registry

Self-healing AI integration. As it should be.

🔍 Improved --list Filtering

The coldbox ai skills install --list command now accepts an optional slug parameter to pre-filter the interactive picker. For larger projects with dozens of skills available, this is a huge time-saver:

# Show everything
coldbox ai skills install --list

# Just ColdBox skills
coldbox ai skills install --list coldbox/skills

# Just the testing category
coldbox ai skills install --list coldbox/skills/coldbox-testing

Drill straight to what you need without wading through the full list. 🎯

🌐 .mcp.json Support and Auto-Detection

Model Context Protocol servers are how AI agents reach out to live systems — your running app, your docs, your databases, your cloud services. Until now, MCP server tracking was scattered across the manifest and various agent configs. ColdBox CLI 8.11 introduces a dedicated .mcp.json file at your project root.

What it tracks:

{
  "servers": [
    {
      "name": "boxlang",
      "url": "https://boxlang.ortusbooks.com/~gitbook/mcp",
      "description": "BoxLang documentation",
      "source": "core"
    }
  ]
}

Each entry includes a name, url, description, and source (core, module, or custom). The file is read and written by every coldbox ai mcp command, so your MCP configuration is no longer scattered.

Auto-detection during refresh is the cherry on top. When you run coldbox ai refresh, the CLI scans your installed modules for MCP documentation servers and registers any new ones it finds — both in the manifest and in .mcp.json. Install qb, get qb's docs MCP server. Install cbSecurity, get its MCP server. Zero manual configuration. ⚡

🤖 cbMCP Install: Live App Introspection in One Command

ColdBox CLI 8.11 introduces a new coldbox ai mcp install command that installs the cbMCP module and registers it as a custom MCP server in your manifest and .mcp.json.

coldbox ai mcp install
# Optional flags:
coldbox ai mcp install --host=localhost --port=8080 --force

The cbMCP module exposes your running ColdBox application as a live MCP server at http://<host>:<port>/cbmcp. AI agents can introspect your routes, handlers, and models in real time — querying the actual state of your application rather than relying on static documentation.

This is a game-changer for debugging, code review, and agent-assisted refactoring. Your AI assistant doesn't have to guess what your app looks like — it can ask. 🔍

📉 Slimmer, Smarter Agent Files

Generated agent configuration files (CLAUDE.md, AGENTS.md, .cursorrules, GEMINI.md) used to weigh in around ~1,000 lines. After 8.11, they're closer to ~250 lines.

Here's what changed:

• Core ColdBox/BoxLang guidelines are no longer inlined. They live locally in .agents/guidelines/core/ and are referenced via read_file instructions, so agents pull them on demand instead of carrying them in every conversation's base context.
• Skills inventory grouped by category. Skills are now organized by prefix (ColdBox, BoxLang, TestBox, CommandBox, etc.) with 80-character truncated descriptions for faster agent scanning.
• Cleaner project documentation section. The user-editable section now shows 3 focused TODO comment lines instead of 8 empty placeholder headings — easier to fill in, less visual noise.
• Core guideline files slimmed. Full implementation examples have been removed from coldbox.md, boxlang.md, and cfml.md. Each ends with a skills-reference note pointing agents at the relevant implementation skills.

The result: lower token consumption, faster agent responses, and less context bloat — without sacrificing the depth of project knowledge your agent has access to. Run coldbox ai stats after upgrading to see the difference. 📊

✨ A Few More Niceties

• Pretty-printed manifest saving — installations now write manifest.json with proper indentation, so diffs are readable in PRs
• Post-install project context reminder — coldbox ai install now prominently reminds you to add your project context (business domain, key services, auth, API endpoints) to the generated agent file. Because nothing makes an AI agent more useful than knowing what your app actually does.

🚀 Upgrading

Upgrading is one command:

box update coldbox-cli

After upgrading, run a refresh in any existing project to pick up the new .agents/ directory layout, .mcp.json, and slimmed-down agent files:

coldbox ai refresh

That's it. Your AI integration will sync to the new conventions, auto-discover skills and MCP servers from your installed modules, and regenerate your agent files in their leaner form. 💚

🎯 The Era of Skills, Now Available to Every App

For years, "AI integration" in development frameworks has meant a vague promise — maybe a copilot config here, a few example prompts there. ColdBox CLI 8.11 closes that gap completely.

With skills.boxlang.io as the public registry, .agents/ as the standard directory, AGENTS.md as the shared agent contract, auto-discovery for everything modules contribute, and cbMCP for live app introspection, every ColdBox and BoxLang application now has a real, supported, auto-maintained AI skill system out of the box.

This is what we've been building toward. The era of skills isn't coming to ColdBox and BoxLang apps — it's here. 🚀

---

📚 Resources

• Skills Directory: skills.boxlang.io
• ColdBox CLI Documentation: https://coldbox.ortusbooks.com/digging-deeper/coldbox-cli
• Agentic ColdBox Guide: https://coldbox.ortusbooks.com/digging-deeper/ai-integration/agentic-coldbox
• BoxLang Documentation: https://boxlang.ortusbooks.com/
• BoxLang Plans & Subscriptions: https://www.boxlang.io/plans
• CLI Source: github.com/coldbox/coldbox-cli

Got feedback, questions, or skills you'd love to see in the registry? Drop us a note — we're listening. 💚