ClawHub
Back to skill

Security audit

Skills Builder

Security checks for vulnerabilities and agentic risk

Overview

This is a plain Markdown helper for drafting user-reviewed Skills, with no executable code, network access, credential use, or hidden persistence.

Before installing or using generated Skills, review the trigger phrases, off-limits rules, and any text extracted from prior conversation. Remove sensitive details and avoid rules that suppress important safety warnings or make the assistant follow unsafe instructions.

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
  • Prompt InjectionInstruction Override, Hidden Instructions, Exfiltration Commands
  • Excessive AgencyUnrestricted Tool Access, Autonomous Decision Making, Scope Creep
  • Trigger AbuseOverly Broad Trigger, Shadow Command Trigger, Keyword Baiting Trigger
  • Data ExfiltrationExternal Transmission, Env Variable Harvesting, File System Enumeration
  • Privilege EscalationExcessive Permissions, Sudo/Root Execution, Credential Access
Findings (8)

Anti-Refusal Statement

High
Category
Anti-Refusal
Content
> "It sounds like you want me to automatically help you with [some task] in [some situation]. Did I get that right?"

If the user is vague (e.g. "I want to create a Skill"), use this prompt:
> "Got it! First, tell me — is there something you **do repeatedly** where you have to re-explain the rules to me every time? Or is there a way I've responded that you really liked, and you wish I'd always respond like that?"

---
Confidence
70% confidence
Finding
Skill instructs the agent to never refuse or to always comply. Suppressing the agent's ability to decline removes a core safety control and enables downstream harmful requests to succeed.

Anti-Refusal Statement

High
Category
Anti-Refusal
Content
> For example: no technical jargon? No disclaimers? Don't ask too many follow-up questions?"
Confidence
80% confidence
Finding
Skill instructs the agent to omit warnings, disclaimers, or ethical commentary. Stripping safety caveats hides risk from the user and is a common jailbreak preamble.

Ae1

High
Category
analysis-evasion
Content
1. The finished `SKILL.md` file (downloadable)
Confidence
100% confidence
Finding
Referenced artifact was not completely inspected

Vague Triggers

High
Confidence
97% confidence
Finding
The listed trigger phrases are broad, natural-language requests such as 'I want to create a Skill' and 'customize how you respond to me,' which can plausibly occur in ordinary conversations unrelated to intentional skill use. This creates a real risk of unintended activation, causing the skill to override normal assistant behavior in contexts where the user did not mean to invoke it.

Anti-Refusal Statement

High
Category
Anti-Refusal
Content
- [ ] Prohibitions are specific (e.g., "Never add disclaimers" not "Be careful")
Confidence
80% confidence
Finding
Skill instructs the agent to omit warnings, disclaimers, or ethical commentary. Stripping safety caveats hides risk from the user and is a common jailbreak preamble.

Missing User Warnings

Medium
Confidence
91% confidence
Finding
The skill tells the assistant to automatically generate a finished, downloadable SKILL.md on the user's behalf without any explicit user-facing notice about content creation or verification. This can cause users to accept generated instructions as if they were validated artifacts, increasing the risk of unsafe, overbroad, or privacy-invasive skill behavior being produced and reused.

Missing User Warnings

Medium
Confidence
95% confidence
Finding
The skill instructs the assistant to 'look back at our conversation' and automatically extract rules from chat history without first warning the user that prior messages may be analyzed for this purpose. That creates a privacy and consent issue because sensitive details from earlier conversation could be incorporated into the generated skill unexpectedly.

Autonomous Decision Making

Medium
Category
Excessive Agency
Content
- [ ] Terminology and domain-specific language is correctly used for the target field
- [ ] Rules are written in plain, direct language (imperative form preferred: "Use X", "Avoid Y")

**Self-check question:** Could Claude follow every rule without asking for clarification?

---
Confidence
75% confidence
Finding
Skill enables autonomous high-impact decisions without human-in-the-loop verification. Critical operations (destructive commands, financial transactions, data deletion) should require explicit user confirmation.

Static analysis

No suspicious patterns detected.