The Agent Skills Directory

[REMOTE_CODE_EXECUTION]: The architectural report generated by this skill configures the Mermaid.js library with securityLevel: "loose". This configuration explicitly disables the library's security sandbox, enabling the execution of JavaScript or malicious HTML tags if they are included in the diagram definitions.
[INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted data from the analyzed codebase, including source code, commit history via git log, and local documentation. Because this content is interpolated into the HTML report and rendered using the insecure Mermaid setting, it creates an XSS (Cross-Site Scripting) vulnerability where a malicious project could execute code on the user's local system. Mandatory Evidence Chain: (1) Ingestion points: Codebase files, git commit history, CONTEXT.md, and ADR files. (2) Boundary markers: No markers are used to isolate codebase content from the report logic. (3) Capability inventory: Executes git commands, writes to the OS temp directory, and triggers browser execution via xdg-open/open/start. (4) Sanitization: No sanitization of project data is performed before rendering.
[COMMAND_EXECUTION]: The skill executes several system commands including git log --oneline to analyze commit hotspots and platform-specific opening commands (xdg-open, open, start) to display the generated HTML review.
[EXTERNAL_DOWNLOADS]: The generated report downloads styling and diagramming assets from well-known services: Tailwind CSS from cdn.tailwindcss.com and Mermaid.js from the JSDelivr CDN (cdn.jsdelivr.net).

improve-codebase-architecture