AI Skills

A skill is a folder an AI coding agent loads when a task needs it: a SKILL.md file with the skill's name, a description the agent reads to decide when the skill applies, and instructions; and reference files the agent opens only when the instructions send it there. The format is the open Agent Skills specification.

The knowledge base is reference material; the skill is a way of working. arx-sdk tells the agent to find the project's ARX.SDK provider from its package reference, to read that provider's guide and types before writing code, to start from the closest verified sample, to use only types and members the types file lists, and to build and check its work before it reports done. Its references are the same files as the MongoDB knowledge base, generated from the library and its samples, so the agent writes the code these pages show.

It targets ARX.SDK 1.0.0-rc.1: SKILL.md carries the version in its metadata (arx-sdk-version), and tells the agent to say so when the project references another one. Downloading it installs nothing: copy it into a folder your agent loads skills from, as below. To read it first, open SKILL.md.

The files the MongoDB part of the skill is made of. The other providers have the same files in their own folders.

Every provider's operation file holds all of its samples, each scenario's code with the mongosh query that checks it, so the skill has no separate examples folder: the agent starts from these.

Claude Code

Claude Code loads skills from .claude/skills/<name>/ in the project (commit the folder and your team gets it too) or from ~/.claude/skills/<name>/ for every project of yours. From the project's folder:

curl -L -o arx-sdk.zip https://arx.migslab.com/ai/skills/arx-sdk.zip
mkdir -p .claude/skills
unzip -o arx-sdk.zip -d .claude/skills
rm arx-sdk.zip

Invoke-WebRequest https://arx.migslab.com/ai/skills/arx-sdk.zip -OutFile arx-sdk.zip
Expand-Archive arx-sdk.zip -DestinationPath .claude/skills -Force
Remove-Item arx-sdk.zip

For every project of yours, unpack it into ~/.claude/skills instead. Claude Code then uses the skill when a task involves ARX.SDK, and /arx-sdk invokes it directly. The AI Instructions page has a CLAUDE.md section that names the project's provider.

Other coding agents

Agents that support the Agent Skills format load the same folder from their own skills directory: unpack the zip there, as their documentation says. For an agent without skill support (or alongside one), keep the folder in the repository and point its instruction file at it; the AI Instructions templates for AGENTS.md, GitHub Copilot and Cursor expect it in docs/arx-sdk/:

curl -L -o arx-sdk.zip https://arx.migslab.com/ai/skills/arx-sdk.zip
mkdir -p docs
unzip -o arx-sdk.zip -d docs
rm arx-sdk.zip

• One version at a time. When the project moves to a newer ARX.SDK, download the skill again and replace the folder, then update the version in your instruction files.
• Don't edit the references. They are regenerated from the library, its guides and its verified samples with every release of these pages; a hand edit is lost, and can disagree with the API.
• Project rules go elsewhere. Conventions of your own belong in your instruction files, not in the skill's folder.

ARX.SDK's model, rules and workflow are the same for every database; what differs is each provider's rules, types, samples, data types and errors, which are in its own folder under references/. One skill means one name for the agent to match, whichever databases a project uses, and a new provider is a new folder, not a new skill.