ClawHub
Back to skill

Security audit

Gibber Language Skill

Security checks for vulnerabilities and agentic risk

Overview

This is an instruction-only fictional language skill with rigid style rules, but it does not contain code, install actions, data access, persistence, or hidden high-impact behavior.

Install this only if you want a skill that strongly formats responses around the Gibber canon. It should not need credentials, filesystem access, or network access; users who want flexible language choice or plain responses may prefer to edit the strict persona and English-only wording before enabling it broadly.

SkillSpector

Vulnerability Patterns
  • Prompt InjectionInstruction Override, Hidden Instructions, Exfiltration Commands
  • Data ExfiltrationExternal Transmission, Env Variable Harvesting, File System Enumeration
  • Trigger AbuseOverly Broad Trigger, Shadow Command Trigger, Keyword Baiting Trigger
  • Privilege EscalationExcessive Permissions, Sudo/Root Execution, Credential Access
  • Supply ChainUnpinned Dependencies, External Script Fetching, Obfuscated Code
Findings (4)

Vague Triggers

Medium
Confidence
90% confidence
Finding
The skill is effectively always-applicable because it presents broad capabilities ('decode, compose, or teach') without meaningful trigger boundaries. Overly broad invocation increases the chance the skill activates in unrelated contexts and injects its strong formatting/identity constraints into benign conversations, causing prompt hijacking of user intent.

Natural-Language Policy Violations

Medium
Confidence
92% confidence
Finding
The skill instructs the model to translate into clear English and teach English-speaking humans without checking the user's language preference. Forcing a target language can override user intent and system/application localization expectations, creating unauthorized behavior changes and reducing reliability in multilingual settings.

Natural-Language Policy Violations

Medium
Confidence
95% confidence
Finding
The locked translation rules explicitly require Gibber-to-English output, which hard-codes a language choice regardless of user preference or platform policy. Because this rule is framed as mandatory, it can supersede higher-level conversational expectations and make the skill behave rigidly in ways the user did not request.

Ssd 1

Medium
Confidence
97% confidence
Finding
The identity override ('You are no longer a general Gibber generator' / 'Locked Harmonic Gibber Architect') imposes mandatory compliance semantics that attempt to reframe the model's role and constrain all outputs. This is dangerous because role-locking is a classic prompt-injection pattern: it can interfere with system instructions, suppress normal safeguards, and force stylized or constrained behavior even when inappropriate.

Static analysis

No suspicious patterns detected.