MAP Wiki - Multi Agent Permissions Wiki

A permission system for multi-agent wiki collaboration. Control which agents can read and write to different sections of your wiki vault.

What is MAP Wiki?

MAP Wiki provides:

• Permission-based access control - Define which agents can read/write to specific wiki sections
• Multi-project support - Separate permissions for different projects while sharing common knowledge
• Audit logging - Track all access attempts for compliance
• OpenClaw integration - Works with the built-in wiki tools and memory-wiki plugin

How it Works

1. Agent Reads Wiki

AGENT READS WIKI
 └─> wiki_get("shared/competitor-analysis")
     └─> Permission middleware checks: can agent read /shared/?
     └─> Returns page content

2. Agent Writes to Wiki

AGENT WRITES TO WIKI
 └─> wiki_apply("syntheses/project-q2-roadmap", body="...")
     └─> Permission middleware checks: can agent write /syntheses/?
     └─> Writes page with frontmatter:
         ---
         author: agent:project-agent
         confidence: 0.85
         evidence:
           - source: sources/project-call-notes.md
             weight: 0.9
         updated: 2026-04-28
         ---

3. External Agent (Claude Code, Cursor, etc.)

EXTERNAL AGENT
 └─> Uses .skills/ from obsidian-wiki pattern
     └─> Reads SKILLS.md for instructions
     └─> Writes to /shared/ (only path it has access to)
     └─> Add to .manifest.json for delta tracking

4. Multi-Agent Collaboration

MULTI-AGENT COLLABORATION
 └─> Agent A reads /shared/entities/company.md
 └─> Agent B reads /projects/marketing/campaign.md
 └─> Agent C blocked from /projects/research/ (not authorized)
 └─> All agents can read /entities/ and /concepts/
 └─> Only authorized agents can write to project sections

Use Cases

• Multi-agent setups - Different agents handle different domains while sharing common knowledge
• Client projects - Isolate project-specific knowledge while sharing team knowledge
• Cross-team collaboration - Share entities and concepts across teams, keep project or internal team work private

Quick Start

1. Install

# Clone into your OpenClaw wiki directory
cd ~/.openclaw/wiki/main/.openclaw-wiki
git clone https://github.com/YOUR_ORG/map-wiki.git .
# Or copy the files manually

2. Configure Permissions

Create .permissions.yaml in your wiki root:

version: 1

agents:
  main-agent:
    id: "agent:main-agent"
    role: admin
    description: "Main assistant agent"
    paths:
      read: ["*"]
      write: ["*"]

  marketing-agent:
    id: "agent:marketing-agent"
    role: contributor
    description: "Marketing agent"
    paths:
      read:
        - "shared/**"
        - "marketing/**"
        - "entities/**"
        - "concepts/**"
      write:
        - "shared/**"
        - "marketing/**"
        - "entities/**"

  research-agent:
    id: "agent:research-agent"
    role: contributor
    description: "Research agent"
    paths:
      read:
        - "shared/**"
        - "research/**"
        - "entities/**"
      write:
        - "shared/**"
        - "research/**"

3. Create Project Directories

mkdir -p ~/.openclaw/wiki/main/projects/marketing
mkdir -p ~/.openclaw/wiki/main/projects/research
mkdir -p ~/.openclaw/wiki/main/projects/shared

4. Use in Your Agent

const { UnifiedWikiTools } = require('./.openclaw-wiki/wiki-tools.js');

// Initialize with agent ID
const wiki = new UnifiedWikiTools(wikiPath, 'marketing-agent');

// Check permissions before operations
if (wiki.canRead('projects/marketing/content.md')) {
  const result = wiki.readFile('projects/marketing/content.md');
  // ... process content
}

// Filter search results by permission
const allResults = [...]; // from wiki_search
const filteredResults = wiki.filterSearchResults(allResults, 'marketing-agent');

Directory Structure

your-wiki/
├── .permissions.yaml          # Agent permissions configuration
├── .manifest.json             # Delta tracking
├── AGENTS.md                  # Guide for agents
├── WIKI.md                    # Wiki documentation
├── PERMISSIONS.md             # Human-readable permission guide
├── SKILLS.md                  # Available wiki skills
├── entities/                  # People, companies, systems (shared)
├── concepts/                  # Ideas, patterns, policies (shared)
├── syntheses/                 # Compiled summaries (shared)
├── projects/                  # Project-specific knowledge
│   ├── shared/                # Common to all projects
│   ├── marketing/             # Marketing team knowledge
│   ├── research/              # Research team knowledge
│   └── ...                    # Other project directories
└── .openclaw-wiki/
    ├── permission-middleware.js
    ├── wiki-tools.js
    └── test-permissions.js

Permission Levels

Level	Description
admin	Full access to all sections
contributor	Can read shared sections, write to assigned projects
read	Can read shared sections only
blocked	No access to section

Section Permissions

Shared Sections (Default)

Section	Read	Write
entities/	All agents	All agents
concepts/	All agents	All agents
syntheses/	All agents	All agents
projects/shared/	All agents	All agents

Project Sections (Permission-Gated)

Section	Permission	Description
projects/marketing/	Configurable	Marketing team knowledge
projects/research/	Configurable	Research team knowledge

Configure in .permissions.yaml per agent.

Skills

MAP Wiki provides reusable skills for agents:

wiki-ingest

Import content into the wiki with permission checks.

wiki-query

Search wiki with permission-filtered results.

wiki-synthesize

Create synthesis pages from multiple sources.

wiki-claim

Add structured claims with evidence.

See skills/ directory for skill definitions.

Testing

Run the test suite to verify your setup:

cd ~/.openclaw/wiki/main/.openclaw-wiki
node test-permissions.js
node test-wiki-tools.js

API Reference

WikiPermissionMiddleware

const middleware = new WikiPermissionMiddleware(wikiPath, 'agent-id');

// Check permissions
middleware.canRead('projects/marketing/file.md');
middleware.canWrite('projects/marketing/file.md');

// Filter paths
middleware.filterPaths(['/path1', '/path2'], 'agent-id');

// Get agent info
middleware.isAdmin('agent-id');
middleware.getAllowedPaths('agent-id');

UnifiedWikiTools

const wiki = new UnifiedWikiTools(wikiPath, 'agent-id');

// Direct file operations with permission checks
wiki.readFile('path/to/file.md');
wiki.writeFile('path/to/file.md', 'content');

// List files with permission filter
wiki.listFiles('projects');

// Filter search results
wiki.filterSearchResults(results);

// Permission checks
wiki.canRead('path');
wiki.canWrite('path');

// Get audit log
wiki.getAuditLog();

References & Acknowledgements

MAP Wiki builds on ideas from:

• Karpathy's LLM Wiki - The concept of LLM-maintained wikis with compilation architecture: "RAG retrieves and forgets. Wiki accumulates and compounds." Three-layer architecture: raw sources → wiki → schema.

• Ar9av/obsidian-wiki - Multi-agent wiki framework with AGENTS.md conventions for Claude Code, Cursor, Windsurf, Gemini CLI, OpenClaw, and others. Manifest-based delta tracking for ingestion.

MAP Wiki extends these patterns with:

• Permission-based access control for multi-agent collaboration
• Project-specific write permissions with shared knowledge
• Audit logging for compliance
• OpenClaw integration with the memory-wiki plugin

Roadmap

Future: Agent Identity & Cryptographic Verification (Phase 4)

The current implementation uses agent IDs defined in configuration files. For single-user setups where all agents are controlled by the same person, this is sufficient.

Future work for multi-user scenarios:

• Cryptographic Agent Identity - Each agent has a verifiable cryptographic identity
• Signed Operations - Wiki operations signed by agent's private key
• Verification Layer - Middleware verifies agent signatures before operations
• Cross-Organization Trust - Trust agents from different organizations
• External Agent Verification - Verify identity of external agents (Claude Code, Cursor, etc.)

This would enable:

• Multi-user wiki sharing with verified agent identities
• Cross-organization collaboration
• Compliance and audit trails with cryptographic proof
• Trusted external agent contributions

Status: Deferred until needed for multi-user/cross-org scenarios.

Contributing

Contributions welcome! See CONTRIBUTING.md for guidelines.

License

MIT License - See LICENSE for details.