ClawHub
Back to skill

Security audit

Distill Self-Improving

Security checks for vulnerabilities and agentic risk

Overview

This skill is a user-directed local file indexing and summarization workflow with disclosed writes to a local knowledge directory and no hidden background, network, credential, or original-file mutation behavior.

Install only if you want a local persistent knowledge index under ~/.agent-knowledge/. Use it on clearly scoped directories or files, and avoid selecting sensitive folders unless you are comfortable with file paths, metadata, and derived summaries being retained locally.

SkillSpector

Vulnerability Patterns
  • 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
  • Supply ChainUnpinned Dependencies, External Script Fetching, Obfuscated Code
Findings (1)

Tp4

High
Category
MCP Tool Poisoning
Confidence
96% confidence
Finding
The skill advertises a content-distillation workflow but also references validator/tooling behaviors (`describe`, `explain`, `self-test`) that can create, copy, modify, and delete files, which materially exceeds a user's expectation of passive knowledge organization. This kind of description-behavior mismatch is dangerous because users and policy systems may approve the skill under a narrower trust model, while the actual available operations include broader filesystem effects and unrelated command functionality.

Static analysis

No suspicious patterns detected.