ClawHub
Back to skill

Security audit

html-collab

Security checks across malware telemetry and agentic risk

Overview

The skill is a coherent HTML review workflow, but it is broad enough to take over generic drafting requests and can overwrite annotated files without a backup.

Install only if you intentionally want generic document drafting to use this HTML review format. Before revision, keep your own copy of the original file, and do not run the image-compression script or paste sensitive annotated documents into chat unless you are comfortable with that content being retained in the conversation.

SkillSpector

By NVIDIA
Vulnerability Patterns
  • Data ExfiltrationExternal Transmission, Env Variable Harvesting, File System Enumeration
  • Excessive AgencyUnrestricted Tool Access, Autonomous Decision Making, Scope Creep
  • Trigger AbuseOverly Broad Trigger, Shadow Command Trigger, Keyword Baiting Trigger
  • Prompt InjectionInstruction Override, Hidden Instructions, Exfiltration Commands
  • Privilege EscalationExcessive Permissions, Sudo/Root Execution, Credential Access
Findings (4)

Context-Inappropriate Capability

Medium
Confidence
87% confidence
Finding
The skill directs the agent to execute a local Python/Pillow script on embedded image data during document reading, which expands behavior from parsing/revision into arbitrary local tool execution. Even though the sample script is bounded to image compression, normalizing code execution from untrusted document content increases attack surface and can lead to unsafe handling of adversarial payloads or environment-dependent abuse.

Vague Triggers

High
Confidence
96% confidence
Finding
The trigger conditions are extremely broad and cause this skill to activate for generic requests like 'write a doc' or 'draft something for review' without requiring explicit user opt-in. That creates prompt-scope hijacking risk: the skill can silently change output format, file-writing behavior, and downstream processing in contexts where the user did not intend to use this workflow.

Missing User Warnings

Medium
Confidence
92% confidence
Finding
The skill instructs overwriting the original file during revision, which is a destructive action performed without requiring confirmation or creating a backup. In collaborative review flows, this can permanently erase prior state, comments, or user-authored content if the revision is wrong or the file was not the intended target.

Context Leakage

High
Category
Data Exfiltration
Content
---

## READ — Extract context from an annotated file

**When:** User provides a `.html` file that has been annotated by a human.
Confidence
89% confidence
Finding
The READ workflow requires the agent to paste the inner HTML of the article and human feedback into the conversation before doing anything else, creating a forced exfiltration path from file contents into chat history. Because conversation logs are durable and may have broader visibility than the file itself, sensitive document contents, annotations, or embedded data can be exposed beyond the minimum needed for processing.

VirusTotal

63/63 vendors flagged this skill as clean.

View on VirusTotal

Static analysis

No suspicious patterns detected.