ClawHub
Back to skill

Security audit

Prompt Debugger

Security checks for vulnerabilities and agentic risk

Overview

This is a static prompt-improvement guide with no executable code, persistence, credential access, or hidden data movement.

Installers should understand that this skill may activate for general prompt-improvement requests and may suggest reasoning-heavy prompt patterns; review prompts before sharing sensitive content and prefer concise justification wording when hidden reasoning disclosure is not desired.

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
  • Trigger AbuseOverly Broad Trigger, Shadow Command Trigger, Keyword Baiting Trigger
  • 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)

Vague Triggers

Medium
Confidence
94% confidence
Finding
The trigger list includes broad phrases such as 'improve this prompt' and 'why isn't this prompt working,' which are common conversational requests and can cause unintended skill activation. Over-broad invocation increases the chance that the skill intercepts unrelated tasks, potentially exposing sensitive prompt content or altering routing behavior unexpectedly.

Intent-Code Divergence

Medium
Confidence
90% confidence
Finding
The Notes section states "Does not execute prompts — analyzes structure and content statically" at L226, but the documented `compare` capability says to A/B test prompt variations and the manifest description says to "compare model responses," which presumes actually running prompts against a model. These documented behaviors conflict at the intent level rather than merely omitting implementation detail.

Natural-Language Policy Violations

Medium
Confidence
94% confidence
Finding
The 'Chain of Thought' pattern explicitly instructs the model to 'Show your reasoning for each step,' which pressures disclosure of internal reasoning without user opt-in. This can lead to unnecessary exposure of hidden reasoning traces, policy-related internal deliberation, or sensitive intermediate analysis patterns when safer summary-style explanations would suffice.

Static analysis

No suspicious patterns detected.