ClawHub
Back to skill

Security audit

AI Watermark Margin

Security checks for vulnerabilities and agentic risk

Overview

The skill performs its advertised image edits, though an optional cropping setting can exceed its stated safety limit.

Keep originals, save results under different filenames, and preview cropping with --report. Do not rely on --max-percent to constrain --fallback. Only upload images you are authorized to share with your chosen AI editor.

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
  • 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 (2)

The workflow instructs uploading the source image to an AI editor without a privacy warning.

Medium
Category
Output Handling
Confidence
95% confidence
Finding

The workflow explicitly directs the user to send the padded image to an AI editor or generator, including online services described earlier in the file. It does not warn that this shares image contents and potentially embedded metadata with an external provider; the existing compliance note addresses AI labeling, not privacy.

Content

No source excerpt is available for this finding.

The fixed-bottom fallback bypasses the documented per-side cropping safety cap.

Medium
Category
Excessive Agency
Confidence
99% confidence
Finding

The fallback replaces the capped detection result with a bottom crop derived directly from --fallback, without enforcing --max-percent. This can remove more image content than the documented safety limit permits; impact is limited to local output integrity, although specifying the source as the destination could destroy the original.

Content

No source excerpt is available for this finding.