HN Showcase
Back to browse
GitHub Repository

Zero-Trust runtime authorization for OpenClaw agents. Block prompt injection and unauthorized tool calls before execution.

25 starsTypeScript

Predicate-Claw – Run Time Assurance (RTA) for OpenClaw via Rust Sidecar

by tonyww·Mar 2, 2026·2 points·1 comment
Visit ProjectView on HN

AI Analysis

●●SolidWizardryNiche Gem

Transparent agent guardrail with <25ms latency, but agent security is early and fragmented.

Strengths
  • •Zero-visibility interception—LLM cannot reason about or evade policy layer; fail-closed by default.
  • •Sub-25ms latency and local zero-egress execution preserves real-time responsiveness and data privacy.
  • •Policy-driven blocking (sensitive paths, shell commands, exfil patterns) with concrete attack table.
Weaknesses
  • •OpenClaw adoption is tiny; the agent runtime ecosystem is still consolidating (LangChain, Anthropic agents, others).
  • •Policies are static rules—no learning or behavior adaptation as threat landscape evolves.
Category
Security
Target Audience

AI agent developers using OpenClaw; teams deploying untrusted LLM-driven automations.

Similar To

Anthropic API request interception · LangChain tool validators · Pydantic AI constraints

Similar Projects

Security●●●Banger

OpenClaw skills degrade agent safety

Behavioral safety testing reveals 45 regressions static analysis misses—guardrails provided.

Big BrainWizardryZero to One
shadab_nazar
123mo ago