Skip to content
Get packs

Guide · Agent skill

How to use the Self-Customize agent skill (self-customize)

Updated · written with AI from the project's own documentation and checked against our catalogue data

The self-customize skill is an instruction file that teaches an AI agent running inside a NanoClaw container how to change its own environment when you ask it to. It is useful if you run a NanoClaw agent and want it to extend its own tools, memory or code over time. This guide explains what the skill does, how to install it, how to trigger it with plain requests, and what to check for safety before relying on it.

Self-Customize Agentnanocoai/nanoclaw · MIT

Quick start

  1. 1Get the skill. Download the ZIP for self-customize from this page.
  2. 2Unzip it into the right folder. Unzip into ~/.claude/skills/ for personal use, or into a project's .claude/skills/ folder. Alternatively, upload it under Settings → Capabilities → Skills in the Claude apps.
  3. 3Confirm your runtime has the required tools. Check that your NanoClaw runtime provides create_agent, send_to_agent, install_packages and add_mcp_server, and that an admin approval flow exists.
  4. 4Make a natural request. Describe the feature, tool or behaviour you want in plain language, including the goal.
  5. 5Approve or review. Approve package installs and MCP server additions when asked. Review builder summaries for code changes before relying on them.
  6. 6Test the new capability. Once the container restarts, ask the agent to use the new tool or behaviour and confirm it works.

What is an agent skill and what does self-customize do?

An agent skill is a folder containing a SKILL.md file with instructions that an AI agent loads when a request matches the skill's description. This one, named self-customize, covers requests to add a capability, install a tool, or change how the agent works.

The skill sorts changes by type. Memory and standing instructions can be edited directly with no approval. System packages (apt) and global npm packages go through the install_packages tool and need admin approval. New MCP servers go through add_mcp_server, also with admin approval. Changes to the agent's source code or Dockerfile are delegated to a separate builder agent created with create_agent, so a human can review the result.

  • Direct edits: memory/ or instructions.prepend.md
  • Admin-approved: install_packages (apt or global npm) and add_mcp_server
  • Delegated: source code or Dockerfile changes to a builder agent
  • The composed CLAUDE.md or AGENTS.md is regenerated every spawn and must not be edited

Install the self-customize skill

Download the ZIP for this skill from this page. The skill is distributed as a folder whose only listed file is SKILL.md. To install it for your user account, unzip it into ~/.claude/skills/. To install it for a single project, unzip it into the project's .claude/skills/ folder.

If you use the Claude apps, you can instead upload the skill under Settings → Capabilities → Skills.

  • Personal install: unzip into ~/.claude/skills/
  • Project install: unzip into the project's .claude/skills/ folder
  • Claude apps: Settings → Capabilities → Skills, then upload

Trigger the skill with natural requests

You do not need to name the skill. The skill loads when your request matches its description: adding a feature, installing a tool, or modifying how the agent works. Ask in plain language and state what you want the outcome to be.

Be specific about the goal. The skill's own guidance says that ambiguous requests should be clarified with the user before builders are created or installs are requested.

  • "Can you add a tool for reading RSS feeds?"
  • "Can you transcribe audio?"
  • "Install ffmpeg so you can process voice messages."
  • "Remember that I prefer short summaries."

What happens when the agent makes a change

For an MCP server, the agent searches for an existing server first (the example in the skill points to mcp.so). If a suitable one exists, it calls add_mcp_server with a name, command and args. After an admin approves, the container restarts with the new server wired up, with no image rebuild.

For source code, the agent calls create_agent to make a builder, then send_to_agent with a task describing the files and changes. The builder reports back what changed, and the main agent confirms with you. Builder rules in the skill include minimal scope, a limit of 200 new lines or 150 modified lines per task, no commits or pushes, and no changes to secrets or .env files.

  • Read the builder's summary before relying on the change
  • Large changes are meant to be split into several smaller builder tasks
  • Packages approved through install_packages trigger an automatic image rebuild and container restart

Know what the skill needs and its safety limits

The skill only works inside a NanoClaw runtime. It requires the create_agent, send_to_agent, install_packages and add_mcp_server tools, an admin approval flow for package installs and MCP server additions, and a container image with bun and pnpm available. It runs scripts and uses the network.

The approval step is the main safety gate. The skill assumes an admin reviews package installs and MCP servers. If your environment has no such approval step, do not use this skill as configured. Also note that the agent can change its own code, so it adds moving parts compared with a fixed agent.

  • Requires: NanoClaw runtime tools, admin approval flow, container with bun and pnpm
  • Runs scripts: yes
  • Uses network: yes
  • Builder agents are told never to modify secrets, credentials or .env files

Tips for using self-customize well

Prototype before promoting. The skill suggests trying a change in the agent's workspace first (for example running pnpm install in /workspace/agent/) and only requesting a container-level install once it proves useful.

Use the skill only for changes you want to keep. For a one-off task, the skill says to do the work in the workspace rather than modify the container.

  • Prototype in /workspace/agent/ before requesting a container-level install
  • Skip the skill for one-off tasks
  • Keep each builder request focused with clear acceptance criteria
pnpm install

What people use it for

  • Adding an MCP server so the agent can read RSS feeds, after an admin approves it
  • Installing a system package such as ffmpeg for processing voice messages
  • Saving standing instructions or notes to memory without any approval step
  • Delegating a focused source code change to a builder agent for later review
  • Prototyping an npm dependency in the agent workspace before a container-level install

Frequently asked questions

How do I install the self-customize agent skill?

Download the ZIP from this page and unzip it into ~/.claude/skills/ for your user account, or into a project's .claude/skills/ folder. In the Claude apps, you can upload it under Settings → Capabilities → Skills instead.

What do I need to use the self-customize skill?

You need a NanoClaw runtime with the create_agent, send_to_agent, install_packages and add_mcp_server tools, an admin approval flow for package installs and MCP servers, and a container image with bun and pnpm available.

Is the self-customize skill free?

The skill is listed under the MIT licence, which is the licence of the nanocoai/nanoclaw repository. Check the licence file in the source repository for the full terms.

Does the agent need approval to change itself?

It depends on the change. Memory and standing instructions can be edited directly. System packages, global npm packages and new MCP servers need admin approval. Source code and Dockerfile changes are delegated to a builder agent for review.

Does this skill run scripts and use the network?

Yes. The skill is listed as running scripts and using the network, so review what it will install and which MCP servers it adds before approving them.

Can I use self-customize with an agent that is not running in NanoClaw?

The skill depends on NanoClaw's specific tools such as install_packages and add_mcp_server and on its container setup. Without those, the instructions will not apply, so pick a different skill or a setup that provides those tools.

New agent skills, repos and AI tools — every week

The best new skills and open-source projects, the fastest-growing repos, and one free prompt. Five minutes to read. Unsubscribe anytime.

Our verdict

Use self-customize if you run a NanoClaw agent and want it to extend its own tools over time, with admin approval and builder review for riskier changes. Pick something else if you want a plain, fixed agent or if your environment has no admin approval step for installs and MCP servers, because the safety gate depends on it.

Download the skill