The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill provides direct instructions to clone a repository from an unverified GitHub account (https://github.com/Paidax01/web-to-figma.git). Downloading and running code from untrusted external sources is a high-risk activity as the source code has not been audited for malicious behavior.
[COMMAND_EXECUTION]: The installation steps involve executing shell commands (git clone, cd) to retrieve the extension's code. This requires the user or agent to perform manual setup operations with unverified software.
[DATA_EXFILTRATION]: The manifest.json configuration snippet requests the <all_urls> host permission. This is a highly sensitive permission that allows the extension to read and modify data on any website the user visits. The background.js script implementation also includes a cross-origin fetch proxy that converts remote resources into data URLs, a pattern that can be leveraged to bypass security policies and exfiltrate page content or sensitive data to external servers.
[INDIRECT_PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection due to how it processes webpage content. Ingestion points: The captureElement and extractText functions in capture.js ingest arbitrary text and structural data from any webpage. Boundary markers: The implementation lacks delimiters or instructions for the agent to ignore embedded commands within the captured data. Capability inventory: The skill uses chrome.runtime.sendMessage, fetch, and file-writing capabilities via downloadJSON. Sanitization: There is no evident sanitization or filtering of the extracted DOM content, meaning malicious instructions hidden on a website could potentially influence the behavior of an AI agent later processing the exported JSON file.

web-to-figma-chrome-extension