Back to skill

Security audit

arkcli-models

Security checks for vulnerabilities and agentic risk

Overview

ClawScan rates this skill benign because the results show no clear malicious behavior, while flagging a mismatch between its stated model-discovery purpose and an inference action; the review is medium-confidence and incomplete.

Consider whether model inference and the inconsistent custom-model routing match what you want from a model-discovery skill. Skillspector inspected all eight listed components, but marked its analysis partial because two referenced local paths were missing from the bundled artifact, so the evidence is incomplete.

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
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 (2)

Description-Behavior Mismatch

High
Category
MCP Tool Poisoning
Confidence
99% confidence
Finding

The metadata lookup and model identifier construction are consistent with model discovery. However, the final command performs a chat inference request, which is not included in the declared query, activation, performance-analysis, or read-only deployment-candidate capabilities. This is an additional operational capability, not merely an implementation detail of metadata retrieval.

Content

No source excerpt is available for this finding.

The regression cases require account-owned custom-model queries to use arkcli-models, contradicting the manifest's explicit routing to arkcli-custommodel.

Medium
Category
Excessive Agency
Confidence
99% confidence
Finding

The manifest limits this skill to public foundation-model discovery and explicitly assigns uploaded or fine-tuned custom models to arkcli-custommodel. These cases instead require enumerating account-owned custom models through arkcli models list and filtering locally; the trigger and scoring instructions reinforce this contradictory scope.

Content

No source excerpt is available for this finding.