How to use the claude-mem install skill for Grok Bot agents

What is an agent skill and what does this one do?

An agent skill is a folder containing a SKILL.md file with instructions that an AI agent can load when a task matches the skill's description. This one is named claude-mem-install and is meant for setting up claude-mem on Grok Bot.

Once claude-mem is installed, a background worker watches conversation transcripts for each Grok Bot agent, summarises them, and injects a live index of past memories into the agent's memory folder so it can recall earlier work. The skill states that Grok Bot has no session-start, file-read or tool-use hooks, so the behaviour is limited to transcript watching and MCP calls.

What you need before installing

The skill's source lists these requirements: Node.js with npx available, Grok Bot installed, and a CMEM Pro account with an API key obtained through installer sign-in (the default path). The remote mode additionally needs a CLAUDE_MEM_MCP_TOKEN plugin variable.

The source states that the skill runs scripts and uses the network.

• Node.js with npx available
• Grok Bot installed
• A CMEM Pro account and API key from installer OAuth (default path)
• CLAUDE_MEM_MCP_TOKEN plugin variable (remote mode only)

How to install the skill

Download the ZIP for this skill from our page and unzip it into ~/.claude/skills/ for personal use, or into a project's .claude/skills/ folder to limit it to that project. If you use the Claude apps, you can instead upload the skill under Settings → Capabilities → Skills.

The skill's own default install command for Grok Bot is shown below. It starts a local worker on 127.0.0.1 and configures the CMEM Pro observer.

Choose an install mode

The default mode, with no --provider flag, uses a local worker and the CMEM Pro observer. The interactive installer pre-selects CMEM Pro. The worker runs on 127.0.0.1 and does not restart if it is already healthy. Live memory index files are written to agents/<uuid>/memory/log/zz-claude-mem-inject.md, and you can disable this with CLAUDE_MEM_GROK_BOT_INJECT_ENABLED=false.

An optional host-login mode uses your local Grok login instead. The source says this is explicit opt-in only and is not the user default. Do not pass --provider host unless you specifically want the loopback shim.

A remote mode points at cmem.ai and requires the CLAUDE_MEM_MCP_TOKEN plugin variable.

• Default: local worker plus CMEM Pro observer
• Host login: npx claude-mem install --ide grok-bot --provider host (opt-in)
• Remote: npx claude-mem install --ide grok-bot --runtime server --server-url https://cmem.ai (needs CLAUDE_MEM_MCP_TOKEN)

How to trigger the skill with natural requests

You do not need special syntax. The skill's description says to use it when setting up claude-mem on Grok Bot, so a plain request such as asking your agent to install claude-mem for Grok Bot should match it. Ask in ordinary words about the goal, for example that you want Grok Bot agents to remember earlier sessions.

Once installed, the source says the agent calls MCP session_start_context at the start of a real task to load context.

• "Set up claude-mem on Grok Bot so my agent remembers past sessions."
• "Install claude-mem for Grok Bot using the default CMEM Pro observer."
• "Set up claude-mem for Grok Bot with the remote cmem.ai server."

Safety notes: scripts, network and data

The skill runs npx commands, starts a local background server and writes files into agent folders. It also uses the network. By default, the summaries are produced through a CMEM Pro hosted inference endpoint at cmem.ai, so conversation transcripts are sent to that third-party service. The source does not detail retention or data handling, so check cmem.ai's policies yourself before using the default path with sensitive work.

If you do not want transcripts leaving your machine, this skill's default path is not suitable. Review the commands before running them.

• Runs scripts and starts a local background worker
• Sends transcripts to cmem.ai under the default path
• Writes files into agent memory folders
• Source does not describe retention or data handling

Tips for a smooth setup

Confirm Node.js and npx work in your terminal before starting. Keep the skill's default mode unless you have a reason to change it, because the other modes need extra setup or opt-in. If you do not want the live index, set CLAUDE_MEM_GROK_BOT_INJECT_ENABLED to false.

• Check that npx runs in your terminal first
• Use the default mode unless you need another
• Disable the injected index with CLAUDE_MEM_GROK_BOT_INJECT_ENABLED=false if not wanted