ClawHub
Back to skill

Security audit

meta-autonomous-research-scale

Security checks for vulnerabilities and agentic risk

Overview

The skill is not destructive or exfiltrating, but it asks agents to persist usage history and modify its own instructions without clear user opt-in or controls.

Review before installing. This skill appears locally focused and not malicious, but it is designed to remember usage patterns and preferences and may encourage agents to rewrite the skill file over time. Install only if you are comfortable with local persistence, and inspect or remove learned_patterns.json and any SKILL.md changes if sensitive research topics or preferences are involved.

SkillSpector

Vulnerability Patterns
  • MCP Least PrivilegeUnderdeclared Capability, Wildcard Permission, Missing Permission Declaration
  • 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 (3)

Lp3

Medium
Category
MCP Least Privilege
Confidence
92% confidence
Finding
The skill instructs callers to use `scripts/learner.py` to persist usage outcomes, preferences, and notes to local files, but the metadata declares no permissions. This creates an undeclared file-write capability that can surprise operators, bypass expected trust boundaries, and enable unintended storage of sensitive task content or user data.

Tp4

High
Category
MCP Tool Poisoning
Confidence
89% confidence
Finding
The description emphasizes research orchestration and self-improvement, but the body also includes persistent logging of invocation outcomes, errors, notes, and user preferences. This mismatch can mislead reviewers and users about the actual data-handling behavior, increasing the risk that privacy-sensitive information is stored without informed consent or proper controls.

Missing User Warnings

Medium
Confidence
95% confidence
Finding
The skill states that it automatically reviews usage and accumulates experience, and the memory file records operation counts, error patterns, user preferences, and improvement suggestions, but it does not present a clear privacy warning or opt-in flow. If users include sensitive research topics, personal preferences, or failure notes, this data may be retained locally in ways they do not expect.

Static analysis

No suspicious patterns detected.