Security audit

Smart Scraper

Security checks across malware telemetry and agentic risk

Overview

This scraper appears purpose-aligned, but optional cache use and watch mode can leave scraped page data on disk.

Installers should treat this as a web scraper that may store scraped content locally when --cache or --watch is used. Avoid sensitive, authenticated, internal, or regulated pages unless the workspace storage path and retention are acceptable, and manually remove memory/scraper-cache data when it is no longer needed.

SkillSpector

By NVIDIA

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 (9)

Intent-Code Divergence

Medium

Confidence: 96% confidence
Finding: The documentation gives contradictory instructions about cache behavior, alternately stating caching is enabled by default and that it requires an explicit `--cache` flag. For a skill that persists scraped page contents to disk, this can mislead users about whether potentially sensitive data will be stored locally, causing unintended privacy and data handling exposure.

Intent-Code Divergence

Medium

Confidence: 95% confidence
Finding: The comparison/usage section conflicts on whether caching is opt-out default behavior or only enabled with `--cache`, which undermines informed user consent and safe operator behavior. In a scraping skill, ambiguity around persistence is security-relevant because operators may process confidential or regulated web content assuming it is not retained when it actually is.

Description-Behavior Mismatch

Medium

Confidence: 91% confidence
Finding: The header documentation advertises only extraction/parsing/status modes, but the implementation also supports watch mode and persists snapshots under local storage. In an agent-skill context, undocumented capabilities materially affect trust and user consent because the skill can retain scraped website content over time and monitor changes without that behavior being disclosed in the top-level interface description.

Description-Behavior Mismatch

Medium

Confidence: 94% confidence
Finding: The skill presents itself as a scraper/parser but also writes parsed content to cache files and watch snapshots on disk. That hidden persistence can expose sensitive scraped data, violate least surprise, and increase privacy risk in shared or multi-tenant agent workspaces where users may assume the operation is transient.

Description-Behavior Mismatch

Medium

Confidence: 96% confidence
Finding: The file header documents only extraction, parsing, and status operations, but the implementation also supports persistent watch mode with baseline capture, diffing, and snapshot storage on disk. Undocumented persistence and monitoring behavior is security-relevant because users may invoke the skill expecting one-time scraping while it silently retains historical page content and metadata for later comparison.

Description-Behavior Mismatch

Medium

Confidence: 95% confidence
Finding: Although described as a scraper/parser, the code persists scraped content to cache and snapshot files under the workspace. Storing extracted page data locally can expose sensitive or proprietary content, create unexpected data retention, and expand the impact of compromise on the host system, especially in agent environments handling untrusted or confidential targets.

Missing User Warnings

Medium

Confidence: 87% confidence
Finding: The README advertises disk caching and watch baselines but does not clearly warn that fetched page content and snapshots may be persisted locally. In a scraping context, this can cause operators to unintentionally store sensitive or regulated data from target sites on disk, increasing exposure through local compromise, backups, or shared workspaces.

Missing User Warnings

Medium

Confidence: 94% confidence
Finding: Conflicting cache warnings create unclear disclosure about local persistence of scraped content, weakening meaningful notice before storing titles, paragraphs, links, tables, prices, images, and metadata on disk. This is dangerous because users may scrape sensitive or proprietary content under the false assumption that no local retention occurs, leading to privacy, compliance, or data leakage issues.

Missing User Warnings

Medium

Confidence: 98% confidence
Finding: Watch mode saves full scraped snapshots to disk without a user-facing disclosure at the point of write. Because snapshots contain page content and metadata over time, this creates covert retention of potentially sensitive data and is more dangerous in an agent skill context where users may not inspect filesystem side effects or realize monitoring leaves historical artifacts.

VirusTotal

64/64 vendors flagged this skill as clean.

View on VirusTotal

Static analysis

No suspicious patterns detected.