ClawHub
Back to skill

Security audit

Claw Post

Security checks across malware telemetry and agentic risk

Overview

This social-posting skill is mostly transparent, but it exposes real-account Facebook group search/join/post automation with conflicting limits that users should review carefully before installing.

Install only if you are comfortable giving an agent-mediated service authority to publish through your logged-in browser sessions. Treat posting, Reddit comments, and Facebook group actions as actions on your real accounts; require explicit confirmation for every post/comment and do not allow automated Facebook group joining or search-to-join-to-post workflows.

SkillSpector

Vulnerability Patterns
  • Trigger AbuseOverly Broad Trigger, Shadow Command Trigger, Keyword Baiting Trigger
  • MCP Tool PoisoningHidden Instructions, Unicode Deception, Parameter Description Injection
  • Prompt InjectionInstruction Override, Hidden Instructions, Exfiltration Commands
  • Data ExfiltrationExternal Transmission, Env Variable Harvesting, File System Enumeration
  • Privilege EscalationExcessive Permissions, Sudo/Root Execution, Credential Access
Findings (4)

Description-Behavior Mismatch

Medium
Confidence
91% confidence
Finding
The skill metadata and narrative claim Facebook group posting is limited to groups the user already belongs to, yet the same skill documents group search and join endpoints. This mismatch can enable an agent to expand from user-authorized posting into growth or membership actions the user may not expect, increasing the risk of unauthorized social actions.

Intent-Code Divergence

Medium
Confidence
95% confidence
Finding
The file repeatedly says auto-join is out of scope and should never be done, but then provides direct API instructions for joining groups and checking membership. Contradictory guidance is dangerous because agent implementations often follow available operational instructions, not policy prose, leading to policy bypass and unintended account actions.

Description-Behavior Mismatch

Medium
Confidence
84% confidence
Finding
The manifest description expands the skill’s advertised behavior beyond the narrower user-facing scope by promoting automated Facebook group search, join, and posting. In an agent ecosystem, this matters because discovery metadata can influence when and how the skill is invoked, potentially enabling higher-risk autonomous social actions that users may not reasonably expect from the stated purpose.

Vague Triggers

Medium
Confidence
88% confidence
Finding
The description uses broad automation language such as human-like posting, automated search/join, and browser-driven actions without defining boundaries, exclusions, or consent checkpoints. In a social-posting skill, that ambiguity increases the risk of spam, unwanted impersonation-style behavior, policy violations, or agent-initiated actions on real user accounts without sufficiently clear limits.

VirusTotal

VirusTotal findings are pending for this skill version.

View on VirusTotal

Static analysis

No suspicious patterns detected.