ClawHub
Back to skill

Security audit

Authority Bias

Security checks for vulnerabilities and agentic risk

Overview

This is a non-executable coaching skill for evaluating authority bias, with no hidden commands, persistence, or data access behavior found.

This skill appears safe to install as a reasoning aid. Users should still treat its historical and business examples as guidance to be independently checked for high-stakes decisions, especially because some source claims are cited in prose rather than enforced by tooling.

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
  • Excessive AgencyUnrestricted Tool Access, Autonomous Decision Making, Scope Creep
  • 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)

Autonomous Decision Making

Medium
Category
Excessive Agency
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.

Content

Scanner excerpt · examples/ai-authority-deference-llm-confidence-and-lab-leaders-2024-2026.md (reported line 23)May include surrounding context.

md
1. Identify: two authorities — the confident LLM answer and the lab-leader pronouncement; narrow question = is *this* answer / *this* timeline correct
2. Expertise vs. positional: fluency and fame are positional signals; neither is calibrated domain expertise on the instant question, and the leader is commercially interested
3. Examine reasoning: does a citation open / a number reconcile / a benchmark exist — or is confidence filling the evidentiary gap (fabricated-citation sanctions as the canonical failure)
4. Dysfunction check: "the AI said so" with no verification; strategy anchored to one quoted timeline; suppressed junior observations
5. Cialdini defense: Q1 and Q2 both default to N — discount the signal, demand the evidence, verify before relying
6. Structural counter: human-in-the-loop fact verification, source logging, independent disinterested experts, timeline premortems

Static analysis

No suspicious patterns detected.