moltbot-safe

A minimal, permissioned, sandboxed execution engine for AI agents.

Purpose

moltbot-safe is a minimal, auditable, safety-first execution substrate for agentic systems. It enforces explicit permissions, isolates execution inside a sandbox, validates all actions, and logs every attempt. The system is minimal, predictable, and secure—no autonomous behavior, no implicit capabilities, no hidden side effects.

Architecture

• Engine: Orchestrates action validation, permission checks, sandboxed execution, and audit logging.
• Permission System: Declarative, least-privilege JSON policy per agent.
• Action Schema: Strict validation of agent actions.
• Audit Log: Every action attempt is logged with timestamp and result.
• Sandbox: All agent actions occur in an isolated directory.

Safety Principles

• Minimalism
• Explicitness
• Isolation
• Least Privilege
• Transparency
• Predictability
• Auditability

Contributing

See CONTRIBUTING.md for safe contribution guidelines.

Documentation

• Architecture
• Threat Model
• Roadmap
• Design Philosophy

---

moltbot-safe is maintained to help the community build safer, more reliable agentic systems.

Implements the principle of least privilege.

📦 Action Schema

A standardized format for all actions an agent can request.
Ensures actions are:

• validated
• well-formed
• safe to execute

Prevents malformed or ambiguous operations.

📜 Audit Log

Every attempted action — allowed or denied — is recorded.
This provides:

• transparency
• traceability
• debugging insight
• accountability

Project Structure

project-root/
├── engine/
│   ├── engine.py
│   ├── permissions.py
│   ├── audit.py
│   ├── action_schema.py
│   ├── permissions.json
│   └── audit.log
├── sandbox/
└── README.md

Safety Principles

1. Least Privilege

Agents only receive the minimum permissions required for their tasks.

2. Isolation

All execution occurs inside the sandbox directory, never the real filesystem.

3. Transparency

Every action is logged, including denied attempts.

4. Extensibility

The system is intentionally minimal but easy to extend with new action types, permission models, or sandbox strategies.

Getting Started

1. Clone the repository: git clone https://github.com/YOURNAME/moltbot-safe.git

2. Install Python 3.7+.

3. Review and customize engine/permissions.json.

4. Use the AgentEngine class to execute actions safely.

Roadmap

• Add more granular permission scopes
• Add optional network sandboxing
• Add dry-run mode for previewing actions
• Add example agents that use the engine
• Add test suite for action validation and permission enforcement

Contributing

Contributions are welcome.
Please open issues or pull requests to propose features, report bugs, or improve documentation.

All contributions should align with the project’s safety-first philosophy.

---

moltbot-safe is maintained to help the community build safer, more transparent, and more responsible agentic systems.