The Agent Skills Directory

Command Execution: The skill invokes system utilities like rsvg-convert and Chromium for rendering. While calling external binaries can be a security consideration if inputs are unvalidated, this skill uses fixed command names and processes rendering data in a way that minimizes exposure.
Dynamic Execution: Helper scripts load modules like Pillow or cairosvg at runtime to check for optional features. Dynamic loading is an area for review because it can lead to unexpected code execution, but here it is used only for dependency discovery within the local environment.
Indirect Prompt Injection: The skill parses external HTML reports, which means it ingests untrusted data. This surface is a potential consideration for prompt injection, though the skill rebuilds content into structured formats and sanitizes CSS and SVG to reduce the risk of malicious instructions influencing the output.
Obfuscation: Base64 and Gzip are used to bundle report payloads. While encoding can be used to hide malicious code from static analysis, in this context it is a standard implementation for self-contained artifacts and includes checks to ensure that no unauthorized scripts are embedded.

build-report