ClawHub
Back to skill

Security audit

diet-plateau-decoder

Security checks for vulnerabilities and agentic risk

Overview

This local weight-trend tool has limited access, but its diet recommendations rely on oversimplified calculations.

Treat its labels as rough estimates, not validated diagnoses. Expected loss and reported noise do not control its conclusions, so avoid changing calories solely from its output. The skill explicitly excludes eating-disorder and medical-edema contexts.

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
  • 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
  • Supply ChainUnpinned Dependencies, External Script Fetching, Obfuscated Code
Findings (1)

The advertised expected-rate and trend-versus-noise analysis does not influence plateau diagnosis.

Medium
Category
Excessive Agency
Confidence
98% confidence
Finding

The expected loss rate and trend-to-noise ratio are calculated but never used to select the verdict; diagnosis instead relies on fixed slope thresholds, sample count, and water flags. Consequently, noisy or statistically inconclusive observations can receive a definitive plateau diagnosis and calorie-reduction advice, while the explanation can claim that an expected loss rate is hidden without testing that hypothesis. The health-advice context increases the consequence of this output-integrity defect, although it does not enable code execution or data theft.

Content

No source excerpt is available for this finding.