The Agent Skills Directory

[COMMAND_EXECUTION]: The skill invokes system-level commands such as xdg-open, open, or start to automatically display a generated architectural review report in the default web browser.
[EXTERNAL_DOWNLOADS]: The generated report fetches front-end libraries (Tailwind CSS and Mermaid.js) from trusted, well-known content delivery networks (CDNs) to handle styling and diagram rendering.
[PROMPT_INJECTION]: The skill ingests data from the local codebase (e.g., file names, commit messages, and ADR content) and incorporates it into an HTML report, which represents an indirect prompt injection surface.
Ingestion points: Project file structure, git history (git log), and documentation files (CONTEXT.md, ADR) are read during the initial exploration phase.
Boundary markers: The generated report does not implement explicit delimiters to differentiate between static report structure and dynamic content extracted from the project.
Capability inventory: The skill has the ability to write files to the system's temporary directory and execute shell commands to open them.
Sanitization: No explicit sanitization or escaping mechanisms are described for cleaning project strings before they are rendered into the HTML document.
[REMOTE_CODE_EXECUTION]: The Mermaid diagramming library in the generated HTML report is configured with securityLevel: "loose". This setting allows the execution of JavaScript within the diagrams. In the absence of proper sanitization of the input codebase data, this could be exploited by an attacker who places malicious strings in file names or documentation to trigger cross-site scripting (XSS) in the context of the user's local browser.

improve-codebase-architecture