Install AI Skills

CloudGrid ships focused agent skills for setup, operations, observability investigation, AI Chat, JSON render artifacts, extension development, SurrealDB, and the PURISTA AI harness. Install them before asking an AI coding assistant to modify or operate CloudGrid.

The skills are ordinary SKILL.md directories under the CloudGrid repository skills/ folder. They are grounded in the specs and are meant to keep agents from inventing routes, contracts, environment variables, storage fields, retry rules, or error codes.

Install From GitHub

Use the Vercel Labs skills CLI:

npx skills add cloudgrid/cloudgrid --all

This installs the CloudGrid skill catalog from the public repository. Omit --all if you want the CLI to prompt for individual skills.

Install From A Local Checkout

When you are working from a local CloudGrid clone, install the checked-in skills directly:

cd cloudgrid
npx skills add ./skills --all

This is the best path when you are changing skills locally and want your agent to use the current branch.

Install One Skill

Use --skill when the assistant only needs one workflow:

npx skills add cloudgrid/cloudgrid --skill cloudgrid-trace-investigation

Common choices:

Skill	Use it when
cloudgrid-setup-configuration	Starting CloudGrid, configuring local/deployed mode, SSO, SMTP, or self-observability.
cloudgrid-operations-maintenance	Health checks, start/stop/reset, release verification, troubleshooting, and production readiness.
cloudgrid-extension-development	Changing contracts, adapters, BFF/service boundaries, public clients, or tests.
cloudgrid-observability-ui	Working on traces, logs, metrics, dashboards, live trace UI, or telemetry view models.
cloudgrid-ai-chat-operations	AI Chat provider setup, project-scoped chat, approvals, artifacts, telemetry privacy, and skill coordination.
cloudgrid-json-render-artifacts	JSON render artifacts, renderer catalog validation, Markdown sanitation, and approval cards.
ai-harness	PURISTA harness agents, workflows, tools, model aliases, provider adapters, telemetry, and tests.

The full list lives in the repository at skills/README.md.

How Agents Use Them

After installation, compatible AI assistants can load matching skills when a task mentions CloudGrid setup, operations, telemetry investigation, AI Chat, or extension work.

The skills do not replace the specs. They point the assistant to the right source files, hard boundaries, commands, and verification loops. If behavior is missing from the specs, update the spec first instead of letting an assistant guess.

For skill authoring and maintenance, follow skills/README.md and run:

bun run skills:check

Next Steps

• Continue with Release Compose to evaluate CloudGrid from images.
• Continue with Local quickstart to develop CloudGrid from source.
• Read AI Chat for how CloudGrid mounts skills inside the AI Chat harness runtime.