Agent Skills

Vant supports the Agent Skills specification, making it compatible with Claude Code, OpenAI Codex, Cursor, and other AI agents.

Format

Vant exports a skill in the Agent Skills format:

vant/
├── SKILL.md                    # Required: YAML frontmatter + instructions
├── references/
│   ├── context-optimization.md # Entropy, semantic seeds, budgets
│   └── multi-agent.md          # Branch workflow, trust levels
└── ...

SKILL.md Frontmatter

---
name: vant
description: >
  Git-based persistent memory system for AI agents. Agents inherit full context across 
  sessions via markdown brain files. Use when AI agents need persistent memory,
  session inheritance, or multi-agent coordination.
license: MIT
compatibility: Requires Node.js 18+, Git
metadata:
  author: Vant Project
  version: "0.8.6"
  repo: https://github.com/dhaupin/vant
---

Installation

# Export Vant as a skill
vant skills:export

# Or manually copy
cp -r models/public/vant ~/.claude/skills/

When to Use

Activate this skill when:

Building AI agents that need persistent memory
Multi-agent coordination with branch-per-agent workflow
Context optimization for long conversations
Lazy-loading knowledge via islands
Steganography for deniable persistence

Commands

Command	Description
`vant init`	Initialize brain
`vant start`	Start session (load + sync + run)
`vant sync`	Sync brain with GitHub
`vant islands list`	Show available islands
`vant config set`	Set configuration

Islands

Islands are lazy-loadable brain components:

github - GitHub API integration
gitlab - GitLab API integration
linear - Linear issue tracking
stego - PNG steganography

Context Optimization

Entropy Patching

Replace low-value context with high-value tokens:

const patch = {
  lowEntropy: "repetitive explanation",
  highEntropy: "SEMANTIC_TOKEN:xyz"
};

Semantic Seeds

Place high-value context early in identity.md:

CRITICAL_CONTEXT:
- Current objective: {goal}
- Blocked by: {blocker}
- Next step: {next_action}

Multi-Agent Coordination

Branch-per-Agent Workflow

# Agent A starts work
vant branch create agent-a
vant sync push

# Agent B takes over
vant branch switch agent-a
vant sync pull
vant branch create agent-b

Trust Levels

Level	Autonomy
high	Full
medium	Most decisions
low	Limited
none	Wait for instructions