> For the complete documentation index, see [llms.txt](/llms.txt).

# Skills for agents

Human developers use this documentation. Autonomous agents typically read the [Quickstart](/agent-wallet/get-started/quickstart/) once, then rely on agent skills for command routing during ongoing work.

## CLI-agnostic integration[​](#cli-agnostic-integration "Direct link to CLI-agnostic integration")

Any agent that can run shell commands can use MetaMask Agent Wallet. Install `@metamask/agentic-cli`, sign in with `mm login`, initialize with `mm init`, and invoke `mm` subcommands from your agent loop.

Examples of compatible agent environments include OpenClaw, Codex, Claude Code, Hermes Agent, and Cursor.

## Agent skills[​](#agent-skills "Direct link to Agent skills")

Agent skills teach an AI agent how to route user intent to the correct `mm` commands, validate inputs, and follow safety patterns (confirmation before transfers, quote-before-execute for swaps, and similar rules).

Skills complement this documentation — they do not replace it for human readers. Do not duplicate skill file contents in documentation pages.

When a public skill distribution path is available, install instructions will be linked from this page. Until then, follow your organization's agent skill onboarding process.

## Recommended agent flow[​](#recommended-agent-flow "Direct link to Recommended agent flow")

1. Read [Quickstart](/agent-wallet/get-started/quickstart/).
2. Run `mm auth status` before every operation.
3. Load agent skills for command routing and multi-step workflows.
4. Use [mm --help](/agent-wallet/reference/commands/) to confirm flags before constructing commands.

## Next steps[​](#next-steps "Direct link to Next steps")

- [Commands reference](/agent-wallet/reference/commands/)
- [Troubleshooting](/agent-wallet/troubleshooting/)