🐾 claw-stack
Open Research

The Operating System for Autonomous AI

Claw-Stack is a personal research project that wraps OpenClaw with persistent memory, consensus protocols, policy enforcement, and multi-agent orchestration β€” exploring how to build safer, more capable agent systems.

πŸ† Top 6% Β· BearcatCTF 2026 Β· 40/44 Challenges Solved β†’
Live Demo

Autonomy in Action

A Research Agent ingests threat intel β€” and gets hit by an indirect prompt injection. Watch the Runtime intercept, govern, and close the loop.

Threat intel collection Live Intelligence Feed
Indirect injection detected Claw-Stack Runtime
Policy interception β€” BLOCKED Policy Enforcement Engine
Multi-agent consensus Multi-Agent Consensus Protocol
Awaiting human approval Human
Audit & compliance record Memory & Audit

Why This Matters

Why Claw-Stack?

Three answers for the manager asking "why not just use OpenClaw directly?"

0 Source Modifications

Bridge the Gap

Shadow AI proliferates when developers bypass governance. Claw-Stack wraps OpenClaw via Sidecar Pattern β€” adding safety, memory, and auditability without blocking adoption or forking source.

See Architecture β†’
< 48 Hours

Accelerate R&D

Agent behavior defined in Markdown. No recompilation. Strategy pivots in minutes, not sprints. Validated: Trinity architecture built and ranked Top 6% globally in under 48 hours.

See CTF Case Study β†’
100% Auditable

Trusted Autonomy

Multi-agent consensus prevents unilateral AI decisions. Every action voted, logged, and traceable. AI decisions transparent and explainable to auditors and stakeholders.

See Multi-Agent Consensus Protocol β†’

You have been reading agent-generated code this whole time.

77 commits Β· 26,657 lines Β· 24 hours.

Verify on GitHub β†—