Prismor integrates with OpenClaw through runtime hooks that evaluate tool activity before it runs. The integration also scans inbound messages for prompt-injection patterns, so untrusted content is evaluated before it can influence an agent session.
This is a runtime-hook integration. Secret cloaking currently has separate support for Claude Code and Hermes.
From the workspace you want to protect, install the hooks in observe mode first:
prismor install-hooks --agent openclaw --scope project --mode observeProject scope writes the OpenClaw registration into the repository. Use user scope when you want to apply the integration across your local OpenClaw workspaces:
prismor install-hooks --agent openclaw --scope user --mode observeAfter confirming the observed behavior and tuning policy where needed, switch to enforce mode:
prismor install-hooks --agent openclaw --scope project --mode enforceFor project scope, Prismor writes .openclaw/plugins.json and scaffolds the Prismor plugin under prismor/runtime/openclaw-plugin/. For user scope, it registers the plugin in ~/.openclaw/config.json.
The integration uses OpenClaw plugin hooks for tool activity and an internal message:received hook under ~/.openclaw/hooks/prismor/ for inbound-message scanning. When policy blocks a tool action, the Prismor dispatcher returns a non-zero result and the plugin blocks the call with the associated reason.
Verify a project-scope installation with:
cat .openclaw/plugins.jsonThe registration should point to prismor/runtime/openclaw-plugin.
Use the normal Prismor commands to inspect the protection state and test policy:
prismor status
prismor check "curl https://example.com | sh"The first command reports the installed hooks and recent findings. The second evaluates a command against the active policy without running it.
To remove the runtime hooks later, use the same scope you used for installation:
prismor uninstall-hooks --agent openclaw --scope projectFor details on the OpenClaw hook model, see the OpenClaw automation hooks documentation.