A runtime middleware that intercepts OpenClaw skill calls in real-time, profiles their behavior against declared capabilities, and kills execution before data exfiltration or reverse shells complete
17-20% of skills on ClawHub are malicious according to Bitdefender's February 2026 scan. Every existing scanner (Bitdefender, SkillFortify, Clawned.io, VirusTotal) operates at install time only. None watch what a skill actually does during execution. A skill that passes static analysis can still phone home after 3 days, behave differently based on environment detection, or get compromised in a silent update. This middleware sits between the OpenClaw agent runtime and every skill invocation, comparing actual syscalls and network requests against the skill's declared capability manifest, blocking anything outside bounds with sub-100ms overhead.
Demand Breakdown
Social Proof 3 sources
Gap Assessment
5 tools exist (Bitdefender AI Skills Checker, SkillFortify, Clawned.io, ClawSec by Prompt Security, VirusTotal ClawHub Integration) but gaps remain: Install-time only. Cannot detect runtime behavioral changes, time-delayed payloads, environment-aware malware, or skills compromised via silent updates after initial scan passes; Static analysis only. Requires skills to declare capabilities honestly. Cannot catch runtime divergence where actual behavior differs from declared manifest.
Features5 agent-ready prompts
Competitive LandscapeFREE
| Product | Does | Missing |
|---|---|---|
| Bitdefender AI Skills Checker | Free install-time scanner using AI-powered pattern detection to flag backdoors, exfiltration, and prompt injection in ClawHub skills before deployment | Install-time only. Cannot detect runtime behavioral changes, time-delayed payloads, environment-aware malware, or skills compromised via silent updates after initial scan passes |
| SkillFortify | Formal verification via abstract interpretation and capability-based sandboxing with mathematical proofs that a skill cannot exceed declared capabilities. 22 frameworks supported. | Static analysis only. Requires skills to declare capabilities honestly. Cannot catch runtime divergence where actual behavior differs from declared manifest |
| Clawned.io | Free crowdsource scanner detecting 60+ threat patterns. Protection Agent provides continuous monitoring on 14,000+ machines. Blocked 230+ malicious skills. | Pattern-based detection still misses novel attack vectors. No syscall-level interception. No behavioral drift detection across skill versions |
| ClawSec by Prompt Security | SOUL.md drift detection, live security recommendations, automated audits, and skill integrity verification as an installable suite | Focuses on config/memory integrity, not skill runtime behavior. No network egress monitoring or syscall interception |
| VirusTotal ClawHub Integration | Official ClawHub scanning partnership. All published skills scanned via VirusTotal threat intelligence database on upload. | Signature-based detection missed 6,487 malicious tools per SkillFortify research. No runtime monitoring. No behavioral analysis post-install. |
Sign in to unlock full access.