Back to skill

Security audit

arkcli-infer-endpoint

Security checks for vulnerabilities and agentic risk

Overview

ClawScan rates this skill suspicious because the scanner reports that some endpoint creation instructions do not match the declared purpose, and that a deletion safeguard can be bypassed by the agent.

Consider the suspicious verdict and Skillspector's DO_NOT_INSTALL recommendation before installing. The scan was partial: two referenced paths were missing, so the evidence is incomplete; this explanation does not establish that the artifact is safe.

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
  • Output HandlingUnvalidated Output Injection, Cross-Context Output, Unbounded Output
  • 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
Findings (4)

Description-Behavior Mismatch

High
Category
MCP Tool Poisoning
Confidence
98% confidence
Finding

The supplied command neither uses the declared arkcli infer endpoint path nor performs Endpoint CRUD, lifecycle management, or model-candidate clarification. Its sole visible operation is code-example generation, a materially different purpose not described in the declaration.

Content

No source excerpt is available for this finding.

Description-Behavior Mismatch

High
Category
MCP Tool Poisoning
Confidence
96% confidence
Finding

Although the description permits +deploy for ordinary deployment requests, it explicitly prohibits substituting that workflow for explicit raw creation. The supplied create reference contains only the prohibited substitution. The omitted arguments prevent assessment of model selection or validation behavior.

Content

No source excerpt is available for this finding.

The post-creation workflow includes stopping the endpoint without warning that this can interrupt inference availability.

Medium
Category
Output Handling
Confidence
90% confidence
Finding

The numbered lifecycle-management step presents endpoint stop/start commands as follow-up actions, rather than merely listing available tools. It does not warn that stopping an endpoint can disrupt applications using it or clarify that this step is optional.

Content

No source excerpt is available for this finding.

The claimed agent-resistant deletion gate contradicts the provided agent invocation.

Medium
Category
Excessive Agency
Confidence
98% confidence
Finding

The documentation claims the deletion gate cannot be manufactured by an agent, yet explicitly shows agents how to satisfy it with an inline environment variable. An agent with shell access can therefore self-authorize irreversible deletion using its existing credentials, defeating the claimed independent confirmation boundary.

Content

No source excerpt is available for this finding.