Extensibility

Vant’s brain system follows a public/private pattern - defaults live in models/public/, your customizations go in models/private/.

Brain Locations

When loading, Vant checks private first -> falls back to public.

Agents

Pre-built agents live in models/public/agents/:

# 26 agents
ls models/public/agents/

To customize an agent:

# 1. Copy to your private brain
cp -r models/public/agents/vant-agent-assistant models/private/agents/my-assistant

# 2. Edit the definitions
edit models/private/agents/my-assistant/AGENT.md

Included Agents

And more: backend, frontend, content, designer, emergency, grep, help, integrate, iterate, ops, reliability, sed, seo

Skills

Pre-built skills live in models/public/skills/:

# 70 skills
ls models/public/skills/

To customize a skill:

# 1. Copy to your private brain
cp -r models/public/skills/vant-skill-git models/private/skills/my-git

# 2. Edit the definitions
edit models/private/skills/my-git/SKILL.md

Included Skills

Core:

• api, git, github, shell, docker, npm, python

Auditing:

• audit-ci, audit-deploy, audit-general, audit-ops, audit-qc, audit-qos, audit-reliability, audit-security, audit-seo

Testing:

• test-unit, test-e2e, test-integration, test-contract, test-load, test-fuzz, test-chaos, test-regression, test-smoke, test-snapshot, test-pen

Chain Skills:

• chain-ci, chain-deploy, chain-pen, chain-qc, chain-security, chain-team, chain-test, chain-full-audit

Reviews:

• review-code, review-docs, review-efficiency, review-performance, review-seo

Other:

• aws, browser, css, deno, email, environment, graphql, hat-black, hat-grey, hat-white, help, island, json, kubernetes, linear, markdown, mcp, postgresql, react, redis, regex, sql, supabase, sync, tailwind, typescript, vite, yaml

Chaining Skills

Chain skills compose multiple skills into workflows:

---
name: chain-ci
description: Run CI pipeline
skills:
  - audit-ci
  - test-unit
  - security
...

Use them to build multi-step workflows.

Override Priority

When you create models/private/agents/my-agent/, it takes precedence over models/public/agents/vant-agent-*.

Same for skills: models/private/skills/my-skill/ -> models/public/skills/vant-skill-*

This lets you:

• Extend Vant without forking
• Keep customizations separate
• Update Vant without losing changes

Best Practices

1. Copy, don’t modify - Start from public copies
2. Keep private changes - Store in your private brain
3. Use descriptive names - vant-agent-fork -> my-custom-agent

Related

• brain.md - Brain system internals
• islands.md - Lazy-loading brain modules
• multi-agent.md - Multi-agent coordination