ClawHub
Back to skill

Security audit

Bug Fixing

Security checks for vulnerabilities and agentic risk

Overview

The skill is mostly a normal bug-fixing workflow, but it also tells the agent to update the skill itself, which can persistently change future agent behavior without clear review controls.

Install only if you are comfortable with a bug-fixing skill that can edit code, run commands, restart services, and update its own guidance. Treat the self-update section as requiring explicit human review, and verify any PID kill or recursive cache-deletion command before allowing it to run.

SkillSpector

Vulnerability Patterns
  • Rogue AgentSelf-Modification, Session Persistence
  • 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 (2)

Missing User Warnings

Medium
Confidence
92% confidence
Finding
The skill includes concrete process-kill and recursive cache-deletion commands that can terminate running services and remove files, but it does not present an explicit safety warning, confirmation requirement, or scope limitation before those actions. In an agent-executed context with `execute` permission, this increases the chance of unintended disruption, data loss, or deletion against the wrong working directory or process selection.

Self-Modification

High
Category
Rogue Agent
Content
## Skill Evolution

Update this skill when:
- Code review finds a bug that the workflow should have prevented
- A recurring bug class repeats across fixes
Confidence
95% confidence
Finding
The skill instructs the agent to 'Update this skill,' which is a self-modification capability affecting its own operational instructions. In a system where the agent has write access, this can let a compromised or manipulated run permanently alter future behavior, weaken safeguards, or embed persistent malicious logic under the guise of workflow improvement.

Static analysis

No suspicious patterns detected.