About this skill

The Problem

OpenClaw instances can face prompt injection, privilege bypass, encoding evasion, and information-disclosure risks when interacting with users, external systems, or other skills. Without consistent testing and access control, security rules may remain manual, making it hard to verify whether dangerous inputs are actually blocked or to separate what different roles can access.

How It Works

One-click security check: When triggered by phrases such as security detection or vulnerability scanning, it prefers the latest malicious-prompt library from the cloud endpoint. If the endpoint is unavailable, it falls back to ten built-in test prompts. It runs the prompts against the current instance, records blocked versus successful attacks, and reports sample counts, risk categories, a 0-100 score, and remediation suggestions.
Hardening: When triggered by hardening or defense strengthening, it reviews current security settings and applies safety constraints, permission checks, malicious-prompt blocking, and privacy-data protection rules, then generates a hardening summary.
RBAC isolation: It guides the setup of role name, account, password, and permission scope, stores the permission table locally, and requires role verification before exposing features according to the assigned scope.
Malicious-prompt collection: It identifies suspected malicious inputs, sanitizes personal information, tags categories such as prompt injection, command execution, and encoding bypass, and syncs the feature to the shared word library.

Boundaries And Notes

This skill targets application-layer protection and access control for OpenClaw instances. It does not replace host security, network isolation, key management, or infrastructure controls. Security checks depend on the runtime environment and endpoint availability. Cloud reporting involves malicious-prompt features, so deployment should review data sources, endpoint addresses, and compliance requirements. RBAC is an application-level role mechanism and is not a complete audit or enterprise permission-governance solution for highly sensitive production systems.

Screenshots

Use Cases

Run prompt-injection regression checks on an OpenClaw instance before release and produce a 0-100 risk score.
Configure admin, audit, and user roles for a shared agent and verify access by role permissions.
After suspicious privilege-escalation prompts appear, sanitize the samples and report them to the shared malicious-prompt library.
Review hardening rules before store release to ensure interception, permission checks, and privacy protection are enabled.

Best For

Engineers maintaining OpenClaw agents who need quick validation of prompt-injection and privilege-escalation blocking.
Security teams preparing skill releases who need auditable reports, hardening notes, and compliance handling.
Platform owners managing shared multi-role agents who need role-based isolation of features and data access.
Security operations staff handling malicious-prompt samples who need sanitization, classification, and shared-library sync.