---

name: vm0 description: Gives the agent the ability to create other agents or update itself. Use this skill when the agent needs to modify its own vm0.yaml (skills, instructions, environment variables) and redeploy, or spin up a new agent from scratch. vm0_secrets:

• VM0_TOKEN

---

VM0 Agent Self-Management

Use this skill to inspect and update the current agent's own configuration — its skills, instructions, and environment variables — by cloning, editing, and redeploying via vm0 compose.

Official docs: https://docs.vm0.ai

---

When to Use

Use this skill when you need to:

• Understand the current agent's vm0.yaml configuration
• Add or remove skills from the current agent
• Update the agent's instructions file
• Change environment variables or secrets references
• Redeploy the agent after configuration changes

---

vm0.yaml Format

Every agent is defined by a vm0.yaml file:

version: "1.0"

agents:
  my-agent:
    framework: claude-code
    instructions: AGENTS.md
    skills:
      - https://github.com/vm0-ai/vm0-skills/tree/main/github
      - https://github.com/vm0-ai/vm0-skills/tree/main/slack
    environment:
      SOME_VAR: "literal-value"
      API_KEY: "${{ secrets.API_KEY }}"
      DEBUG: "${{ vars.DEBUG }}"

Fields

Field	Description
version	Config version — always "1.0"
agents	Map of agent definitions (key = agent name)
framework	Agent runtime — use claude-code
instructions	Path to the instructions/system-prompt file (e.g. AGENTS.md)
skills	List of skill URLs to inject into the agent
environment	Environment variables available at runtime

Variable Syntax

Syntax	When to use
"literal-value"	Plain non-sensitive config
"${{ secrets.NAME }}"	Sensitive values (API keys, tokens)
"${{ vars.NAME }}"	Non-sensitive remote variables

Important: Each skill's SKILL.md declares its own required secrets and vars in the frontmatter (vm0_secrets / vm0_vars). These are automatically injected when the skill is loaded — do not redeclare them in environment. Only put entries in environment that are specific to this agent and not already covered by a skill's frontmatter.

---

Finding Skills

Skills come from two marketplaces. Search both when looking for new capabilities.

1. skills.sh (33,700+ community skills)

curl -s "https://skills.sh/api/search?q={keyword}"

2. vm0-ai/vm0-skills (70+ curated integrations)

Browse at: https://github.com/vm0-ai/vm0-skills

Priority rule: If a skill exists in both marketplaces, prefer the vm0-ai/vm0-skills version — it is optimized for VM0 agent workflows and has consistent quality.

Converting search results to vm0.yaml URLs

Skills are referenced in vm0.yaml as GitHub tree URLs:

Source	URL format
vm0-ai/vm0-skills	https://github.com/vm0-ai/vm0-skills/tree/main/{skill-name}
anthropics/skills	https://github.com/anthropics/skills/tree/main/skills/{skill-name}
vercel-labs/agent-skills	https://github.com/vercel-labs/agent-skills/tree/main/skills/{skill-name}
Other repos	https://github.com/{owner}/{repo}/tree/main/skills/{skill-name}

Checking required credentials

After picking a skill, read its SKILL.md to see what secrets or vars it needs:

curl -s "https://raw.githubusercontent.com/vm0-ai/vm0-skills/main/{skill-name}/SKILL.md" | head -20

Look for vm0_secrets and vm0_vars in the frontmatter. Store any missing values before deploying:

vm0 secret set SKILL_TOKEN --body "value"

---

CLI: clone & compose

Clone — download current agent config

vm0 agent clone downloads a deployed agent's configuration to the local filesystem. It writes vm0.yaml and the instructions file (e.g. AGENTS.md) to a local directory.

# Clone into a new subdirectory (directory must not already exist)
vm0 agent clone {agent-name} ./agent-config

# Clone into current directory
vm0 agent clone {agent-name} .

What it downloads:

• vm0.yaml — the compose configuration
• The instructions file referenced in vm0.yaml (e.g. AGENTS.md)

Note: Secret values are preserved as ${{ secrets.NAME }} placeholders — the actual values are never written to disk.

---

Compose — deploy updated config

vm0 compose deploys a vm0.yaml to the platform, creating or updating the agent.

# Deploy with confirmation prompt
vm0 compose vm0.yaml

# Deploy non-interactively (skip confirmation)
vm0 compose vm0.yaml -y

If the yaml references secrets that don't yet exist on the platform, compose will prompt to set them (use -y to skip in automated contexts).

---

Self-Update Pattern

An agent can modify its own configuration by combining clone → edit → compose. This lets the agent update itself — adding new skills, changing instructions, or updating env vars — without any human intervention.

1. Clone own config

vm0 agent clone {current-agent-name} /tmp/self-config

2. Inspect current state

cat /tmp/self-config/vm0.yaml
cat /tmp/self-config/AGENTS.md

3. Edit vm0.yaml

For example, add a new skill:

# Edit /tmp/self-config/vm0.yaml to add a skill entry under skills:
#   - https://github.com/vm0-ai/vm0-skills/tree/main/notion

Or update the instructions file:

# Edit /tmp/self-config/AGENTS.md to refine the system prompt

4. Redeploy

cd /tmp/self-config
vm0 compose vm0.yaml -y

The agent is now updated. The next run will use the new configuration.

---

Guidelines

1. Agent name: When self-updating, the agent name comes from the agents: key in vm0.yaml. The current agent name is available in the VM0_AGENT_NAME environment variable if set.
2. Clone target must not exist: vm0 agent clone fails if the destination directory already exists. Use a fresh path like /tmp/self-config or add a timestamp.
3. compose -y for automation: Always pass -y when running compose non-interactively to skip confirmation prompts.
4. Instructions file: The instructions file path in vm0.yaml is relative to the directory where vm0 compose is run. Keep both files in the same directory.
5. Secrets are not cloned: Secret values are never downloaded — only the ${{ secrets.NAME }} reference is preserved. Don't try to read secret values from the cloned config.