ClawHub
Back to skill

Security audit

Self-Renewal

Security checks for vulnerabilities and agentic risk

Overview

This is a low-risk coaching skill, though its insistence on continued challenges can encourage unnecessary belief changes.

Use it as a reflection aid. Keep beliefs that withstand credible evidence, set a limit on repeated challenges, and review suggested behavior changes before acting.

A.I.G

Vulnerability Patterns
  • Skill Instruction HijackingAlters the agent's session goals or safety constraints when the skill loads
  • Agent Memory PoisoningWrites attacker-controlled rules into memory that affect later sessions
  • Remote Payload Retrieval and ExecutionFetches external code whose behavior can change after review
  • Embedded Malicious CodeShips malicious scripts inside the skill and executes them locally
  • Unauthorized Access and Privilege EscalationObtains permissions beyond the task's legitimate needs

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)

The process requires further disconfirmation whenever all beliefs remain unchanged, contradicting the manifest's exclusion of renewal for its own sake.

Medium
Category
Excessive Agency
Confidence
95% confidence
Finding

The stop-rule treats all five beliefs remaining unchanged as proof that the experiment was performed incorrectly, requiring repeated challenges without an evidence-based stopping condition. This can pressure an agent into recommending unsupported belief or strategy changes, or continuing an unproductive loop, despite the stated exclusion of renewal for its own sake. The coaching context and explicit protection of valid beliefs reduce the scope of harm, but the contradictory operational rule undermines those safeguards; this is a workflow-integrity weakness rather than evidence of malicious prompt injection.

Content

No source excerpt is available for this finding.

Static analysis

No suspicious patterns detected.