Back to skill

Security audit

Network AI

Security checks for vulnerabilities and agentic risk

Overview

ClawScan rates this skill suspicious because scanners found confirmed weaknesses that could redirect writes to persistent context, permission, audit, or coordination data when an attacker can manipulate directories.

Consider the confirmed write-redirection weaknesses and the incomplete reference analysis before installing. The verdict is suspicious, a review category; the reported evidence does not establish malicious intent.

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 (3)

The write helper follows directory symlinks despite explicitly promising never to write through a symlink.

Medium
Category
Excessive Agency
Confidence
99% confidence
Finding

The helper checks resolved directory containment but opens the original path, and O_NOFOLLOW protects only the final component. An attacker able to manipulate directories can alias one environment to another within data/, replace a parent after validation, or redirect data/ itself because its resolved target is accepted as the trusted root; a stable environment symlink pointing outside an unchanged root is correctly rejected. In this persistent-memory skill, these weaknesses can redirect context overwrites or audit appends, although exploitation requires filesystem access and remains limited by the process's permissions.

Content

No source excerpt is available for this finding.

The write helper follows directory symlinks despite explicitly promising never to write through a symlink.

Medium
Category
Excessive Agency
Confidence
99% confidence
Finding

The helper follows parent-directory symlinks during os.open despite checking resolved containment, while O_NOFOLLOW covers only the final filename. An attacker with directory-manipulation access can redirect an environment to another environment within data/, race the validation, or redirect data/ itself; stable environment symlinks outside an unchanged root are blocked. Because this script rewrites permission grants and appends audit records, redirection can modify the wrong environment's authorization state or corrupt another writable target.

Content

No source excerpt is available for this finding.

The documented no-symlink write confinement is not enforced for the data directory or parent directories.

Medium
Category
Excessive Agency
Confidence
99% confidence
Finding

Resolving _DATA_ROOT accepts a symlinked data directory as the trusted root, permitting writes outside the intended physical data directory; additionally, parent directories can be replaced with symlinks between validation and os.open(), since O_NOFOLLOW protects only the final path component. An attacker able to modify those directories could redirect state-file truncation or audit-log appends to other writable locations, compromising the shared orchestration and budget records despite the documented confinement guarantee.

Content

No source excerpt is available for this finding.