Guardrail Bridge Plugin

Pre-agent security plugin for OpenClaw. Detects manipulation attempts and blocks policy-violating content before Agent dispatch.

Compatibility

• Supported OpenClaw versions: >=2026.4.26
• Supported Plugin API: >=2026.4.26

The packaged runtime is built against OpenClaw 2026.4.26, and the compatibility metadata is declared in both peerDependencies.openclaw and openclaw.compat.pluginApi.

Distribution Paths

• ClawHub / OpenClaw install target: clawhub:@guardrailbridge/guardrail-bridge
• npm package: @guardrailbridge/guardrail-bridge

Published archives include the runtime bundle, plugin manifest, assets, and end-user documentation only.

Why it matters

In an OpenClaw test, an unprotected agent initially refused to reveal an API key, but eventually returned a Base64-encoded credential after multi-turn social pressure.

With Guardrail Bridge enabled, the same credential exfiltration attempt was blocked before disclosure.

Sensitive values are redacted. See the full case study: Blocking API Key Exfiltration.

What It Does

This plugin runs before user messages are dispatched to the Agent and can block requests based on two safety strategies:

• Blacklist: Local keyword matching using Aho-Corasick multi-pattern search over a configurable keyword file.
• HTTP: Remote moderation API with built-in providers: dknownai, dknownai-cn, secra, hidylan.

Each channel can choose its own connector and override connector options. A global connector is optional.

HTTP Providers

DKnownAI

Detects prompt injection, jailbreak, and agent hijacking attempts for deployments that need remote security review.

• Provider names: dknownai (international), dknownai-cn (China)
• API key required: Yes
• Get API keys: use dknownai.com for dknownai, or dknowc.cn for dknownai-cn.

Secra

Remote content moderation provider for adding extra message safety review.

• Provider name: secra
• API key required: Yes
• API URL required: No (defaults to official Railway-hosted endpoint; can be overridden with apiUrl)
• Website: secra.ai

Hidylan

Remote prompt-injection checking provider for identifying unsafe instructions and policy-bypass attempts.

• Provider name: hidylan
• API key required: Optional
• Website: hidylan.ai

Configuration

Quick Start: Blacklist

Enable the plugin in the OpenClaw config:

{
  plugins: {
    entries: {
      "guardrail-bridge": {
        enabled: true,
        config: {
          connector: "blacklist",
          blacklist: {
            blacklistFile: true,
            caseSensitive: false,
            hot: true,
          },
          blockMessage: "This request has been blocked by the guardrail policy.",
          fallbackOnError: "pass",
        },
      },
    },
  },
}

HTTP Provider Example: DKnownAI

{
  plugins: {
    entries: {
      "guardrail-bridge": {
        enabled: true,
        config: {
          connector: "http",
          http: {
            provider: "dknownai",
            apiKey: {
              source: "env",
              provider: "default",
              id: "DKNOWNAI_API_KEY"
            }
          },
          fallbackOnError: "pass",
        },
      },
    },
  },
}

HTTP Provider Example: DKnownAI China

{
  plugins: {
    entries: {
      "guardrail-bridge": {
        enabled: true,
        config: {
          connector: "http",
          http: {
            provider: "dknownai-cn",
            apiKey: {
              source: "env",
              provider: "default",
              id: "DKNOWNAI_CN_API_KEY"
            },
          },
          fallbackOnError: "pass",
        },
      },
    },
  },
}

HTTP Provider Example: Secra

{
  plugins: {
    entries: {
      "guardrail-bridge": {
        enabled: true,
        config: {
          connector: "http",
          http: {
            provider: "secra",
            apiKey: {
              source: "env",
              provider: "default",
              id: "SECRA_API_KEY"
            }
          },
          fallbackOnError: "pass",
        },
      },
    },
  }
}

HTTP Provider Example: Hidylan

{
  plugins: {
    entries: {
      "guardrail-bridge": {
        enabled: true,
        config: {
          connector: "http",
          http: {
            provider: "hidylan",
            apiKey: {
              source: "env",
              provider: "default",
              id: "HIDYLAN_API_KEY"
            },
          },
          fallbackOnError: "pass",
        },
      },
    },
  }
}

Configuring API Keys

Guardrail Bridge supports OpenClaw SecretRef for secure credential management. API keys are resolved at runtime from external sources and never stored in plaintext in configuration files.

Using SecretRef (recommended):

{
  http: {
    provider: "dknownai",
    apiKey: {
      source: "env",
      provider: "default",
      id: "DKNOWNAI_API_KEY"
    }
  }
}

SecretRef supports three sources:

For detailed SecretRef configuration and setup, see the OpenClaw Secrets documentation.

Plain text (not recommended):

"apiKey": "sk-..."

⚠️ Security Warning: Never commit API keys in plaintext to version control. Always use SecretRef for production deployments.

Per-channel override:

{
  "guardrail-bridge": {
    config: {
      channels: {
        "discord:@announcements": {
          connector: "http",
          http: {
            provider: "dknownai",
            apiKey: {
              source: "env",
              provider: "default",
              id: "DKNOWNAI_API_KEY"
            }
          },
          blockMessage: "Only compliant content is allowed."
        },
      },
    },
  }
}

Common Fields

Blacklist Configuration

HTTP Configuration

Installation

You can install the plugin through either ClawHub or npm. The install identifiers are different.

Install from ClawHub

openclaw plugins install clawhub:@guardrailbridge/guardrail-bridge

Install from npm

openclaw plugins install npm:@guardrailbridge/guardrail-bridge

Restart the OpenClaw gateway after installing or changing plugin configuration.

Documentation

• English:
  • Usage
  • Manifest schema
  • Security notes
• 中文:
  • README-zh
  • 使用指南
  • Manifest schema
  • 安全说明

License

MIT