The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted data from the local codebase (source files, git history, domain glossary) to generate architectural recommendations.
Ingestion points: The agent reads project source files, executes git log --oneline, and parses GLOSSARY.md and ADRs in docs/adr/ to inform its analysis.
Boundary markers: The instructions do not define delimiters or warnings to prevent content from these files from being interpreted as instructions during the interpolation into the HTML report.
Capability inventory: The skill has the ability to write HTML files to the system temporary directory and execute shell commands (xdg-open, open, start) to launch them in a browser.
Sanitization: There is no requirement or logic provided to sanitize or escape strings extracted from the codebase before they are rendered into the HTML report.
[DYNAMIC_EXECUTION]: The skill generates a dynamic HTML report at runtime and executes it by opening it in the user's default browser. The Mermaid diagramming configuration in HTML-REPORT.md explicitly sets securityLevel: "loose". This setting allows the execution of HTML tags and scripts within diagram labels, which presents a Cross-Site Scripting (XSS) vulnerability surface if malicious content exists in the scanned codebase.
[COMMAND_EXECUTION]: The skill uses system commands to perform its exploration and presentation phases. It uses git log to identify active areas of the codebase and platform-specific commands (xdg-open, open, or start) to automatically open the generated report for the user.
[EXTERNAL_DOWNLOADS]: The generated architectural report fetches the Tailwind CSS framework and the Mermaid diagramming library from well-known Content Delivery Networks (CDNs) at runtime.
Sources: The report includes script tags pointing to cdn.tailwindcss.com and cdn.jsdelivr.net.

improve-codebase-architecture