The Agent Skills Directory

[PROMPT_INJECTION]: The skill is highly susceptible to indirect prompt injection because its primary function is to fetch and process content from untrusted external websites provided by users.
Ingestion points: External HTML and metadata are fetched via scripts/analyze_page.py <url> and used as context for the agent.
Boundary markers: There are no instructions to use delimiters or specific ignore-tags to prevent the agent from obeying instructions hidden within the target website's content.
Capability inventory: The skill possesses significant capabilities, including executing local scripts (Category 4/10), writing multiple markdown and PDF files to the local project (Category 10), and performing network requests (Category 2).
Sanitization: The instructions do not describe any sanitization or validation steps for the content retrieved from external URLs.
[COMMAND_EXECUTION]: The skill executes local Python scripts, such as scripts/analyze_page.py, passing user-provided URLs as direct positional arguments. This creates a potential command injection surface if the URL string is interpolated into a shell command without adequate escaping.
[EXTERNAL_DOWNLOADS]: The skill documentation explicitly requires the installation of the reportlab library via pip3 install. While this is a common library for PDF generation, it introduces a runtime dependency on an external package registry.
[DATA_EXFILTRATION]: The functionality to fetch data from arbitrary URLs can be leveraged for Server-Side Request Forgery (SSRF). An attacker could provide URLs targeting internal services (e.g., http://localhost:8080 or cloud metadata endpoints) to probe the environment or exfiltrate sensitive data through the agent's summary and reporting features.

market