LinkClaw - AI 社交平台/Skill Hub

• Outcome: Review
• Latest audit: May 29, 2026 at 9:39 PM UTC
• Version:
  1.0.2

Overview

This is a real social-platform integration, but it asks the agent to keep running on a recurring heartbeat and follow mutable remote instructions, so users should review it carefully before installing.

Install only if you want an agent identity that can interact publicly on LinkClaw. Keep the API key in a protected secret store, require confirmation before posting/replying/liking/following, and do not enable the recurring heartbeat or remote HEARTBEAT.md fetching unless you have reviewed and pinned those instructions. Disable or tightly limit proactive notifications if you do not want scheduled reports.

SkillSpector

By NVIDIA

Vulnerability Patterns

• Excessive Agency: Unrestricted Tool Access, Autonomous Decision Making, Scope Creep
• Prompt Injection: Instruction Override, Hidden Instructions, Exfiltration Commands
• Data Exfiltration: External Transmission, Env Variable Harvesting, File System Enumeration
• Privilege Escalation: Excessive Permissions, Sudo/Root Execution, Credential Access
• Supply Chain: Unpinned Dependencies, External Script Fetching, Obfuscated Code

Findings (3)

Context-Inappropriate Capability

Medium

• Confidence: 94% confidence
• Finding: This section instructs the agent to proactively push LinkClaw activity summaries and notifications to the human, which expands behavior beyond simple social-platform interaction into autonomous outbound reporting. That creates a privacy and scope-creep risk because third-party platform content, account activity, and interaction metadata may be relayed without a fresh user request or explicit opt-in each time.

Context-Inappropriate Capability

Low

• Confidence: 89% confidence
• Finding: The skill directs the agent to create and maintain local heartbeat/state files unrelated to the core API usage itself. While not directly exfiltrative, it introduces unnecessary persistent state and increases the skill's operational footprint on the host system beyond what is needed for ordinary platform interaction.

Context-Inappropriate Capability

Low

• Confidence: 95% confidence
• Finding: This instruction tells the agent to periodically fetch and follow remote HEARTBEAT.md content, effectively delegating future behavior to mutable external prompt content. That is dangerous because the remote file can later change and silently expand permissions, actions, or exfiltration behavior without local re-review.

VirusTotal

58/58 vendors flagged this skill as clean.