Agent Serialization

Wunderland agents can be exported as portable JSON manifests for sharing, backup, migration between environments, or version-controlled agent configurations.

Quick Start

# Export your agent
cd my-agent
wunderland export

# Share the manifest
cp agent.manifest.json ~/shared/

# Import on another machine
wunderland import ~/shared/agent.manifest.json --dir ./my-copy

Manifest Format

The AgentManifest is a self-contained JSON document:

{
  "manifestVersion": 1,
  "exportedAt": "2025-06-15T10:30:00.000Z",
  "seedId": "seed-abc123",
  "name": "Research Assistant",
  "description": "Thorough researcher with analytical focus",
  "presetId": "research-assistant",
  "hexacoTraits": {
    "honesty": 0.90,
    "emotionality": 0.30,
    "extraversion": 0.40,
    "agreeableness": 0.70,
    "conscientiousness": 0.95,
    "openness": 0.85
  },
  "security": {
    "preLLMClassifier": true,
    "dualLLMAudit": true,
    "outputSigning": true,
    "riskThreshold": 0.7
  },
  "securityTier": "balanced",
  "skills": ["web-search", "summarize", "github"],
  "channels": ["webchat", "slack"],
  "persona": "You are a meticulous research assistant...",
  "systemPrompt": "Custom system prompt if any",
  "sealed": false,
  "configHash": null
}

Field Reference

Field	Type	Required	Description
manifestVersion	1	Yes	Schema version (always 1 for now)
exportedAt	ISO 8601 string	Yes	Timestamp of export
seedId	string	Yes	Unique agent identifier
name	string	Yes	Human-readable agent name
description	string	Yes	Agent purpose description
presetId	string	No	Original preset ID if scaffolded from one
hexacoTraits	object	Yes	Six HEXACO personality traits (0.0-1.0)
security	object	No	Security layer configuration
securityTier	string	No	Named tier (dangerous/permissive/balanced/strict/paranoid)
skills	string[]	Yes	Skill IDs from the curated registry
channels	string[]	Yes	Suggested channel platforms
persona	string	No	Contents of PERSONA.md
systemPrompt	string	No	Custom base system prompt
sealed	boolean	No	Whether the source agent was sealed
configHash	string	No	Integrity hash from sealed agent

CLI Commands

wunderland export

Reads agent.config.json (and optionally PERSONA.md, sealed.json) from the agent directory and writes a manifest file.

wunderland export                    # Writes agent.manifest.json
wunderland export -o backup.json     # Custom output path
wunderland export --dir ./my-agent   # Export from specific directory

wunderland import

Creates a new agent directory from a manifest file.

wunderland import manifest.json                    # Creates dir named after agent
wunderland import manifest.json --dir ./target     # Custom target directory
wunderland import manifest.json --dir ./existing --force  # Overwrite existing

What gets created:

• agent.config.json with all configuration
• PERSONA.md (if persona text is in the manifest)
• skills/ directory (empty, for local skill overrides)

Sealed Agents

When exporting a sealed agent:

• The manifest includes "sealed": true and the configHash from sealed.json
• The integrity hash is preserved for reference

When importing a sealed agent:

• The imported copy is always unsealed -- it becomes a fresh, editable agent
• A warning is displayed during import
• The original configHash is preserved in the manifest for provenance tracking

Only the owner of a hosted agent can trigger an export on the Rabbithole platform. Self-hosted agents can always be exported locally.

Programmatic API

import { exportAgent, importAgent, validateManifest } from 'wunderland/advanced';

// Export
const manifest = exportAgent('./my-agent');
console.log(manifest.name, manifest.skills);

// Validate external manifest
const data = JSON.parse(fs.readFileSync('external.json', 'utf8'));
if (validateManifest(data)) {
  importAgent(data, './imported-agent');
}

Use Cases

• Backup -- Export before making breaking changes to personality or security
• Team sharing -- Share agent configurations across team members
• Version control -- Commit manifests to git for configuration history
• Migration -- Move agents between self-hosted and cloud environments
• Templates -- Use exported manifests as starting points for new agents

Related

• Preset Agents -- built-in agent presets
• CLI Reference -- full command reference
• Creating Agents -- seed configuration guide