Back to skill

Security audit

AlgoRover Research

Security checks for vulnerabilities and agentic risk

Overview

The scanners report a research skill with no executable scripts and a disclosed install step, and the verdict is benign because the flagged behaviors were assessed as expected for its purpose.

Before installing, consider whether you are comfortable running the latest ClawHub CLI with your privileges; the scanners note it is not pinned to a reviewed version. Also note the reported deletion instructions and English locale default. The scan covered all three listed components and completed, but its findings do not establish that the mutable CLI or installed skill is risk-free.

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
94% confidence
Finding

The only supplied command performs skill installation, which is materially different from the declared operational purpose. Installation could be a setup step for the described functionality, but the supplied chunk contains no implementation or invocation of that functionality. The behavior of the installed skill cannot be established from this command alone.

Content

No source excerpt is available for this finding.

MCP server referenced without pinned version: 'npx clawhub'.

Medium
Category
MCP Rug Pull
Confidence
96% confidence
Finding

The documented installation command uses npx clawhub@latest, which downloads and executes a mutable CLI release with the user's privileges. Although this is an installer rather than the MCP server itself, a compromised or malicious future release could execute arbitrary code; the legitimate installation context does not eliminate that supply-chain risk.

Content

No source excerpt is available for this finding.

L23 introduces deletion of individual results and all Research data, beyond the manifest's stated research and job-management operations.

Medium
Category
Excessive Agency
Confidence
94% confidence
Finding

The manifest at L03 describes discovery, backtesting, job inspection or cancellation, and report downloads, but L23 additionally instructs the agent to delete results or all Research data. Deleting stored research data is a distinct destructive operation, not an implementation detail of cancelling active jobs; requiring explicit requests and confirmation limits accidental use but does not resolve the scope mismatch.

Content

No source excerpt is available for this finding.

Discovery forces the English locale for users of all languages other than Chinese without opt-in.

Low
Category
Output Handling
Confidence
91% confidence
Finding

The workflow requires locale: "en" for "English or other languages" and preserves that locale in subsequent tools. This imposes an English locale on users of other languages without offering a choice or documenting why the restriction is necessary.

Content

No source excerpt is available for this finding.