The Agent Skills Directory

[COMMAND_EXECUTION]: The skill uses shell commands to explore the project's history (git log) and to open the generated report in the system's default browser using xdg-open, open, or start.
[INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted data from the local codebase, including filenames, commit messages, and documentation contents (CONTEXT.md, ADRs), which are then interpolated into an HTML report.
Ingestion points: Data is pulled from git log output and files throughout the repository as described in SKILL.md.
Boundary markers: No explicit delimiters or boundary markers are used when interpolating codebase data into the HTML scaffold.
Capability inventory: The skill executes subprocesses (git), writes files to the temporary directory, and triggers network requests through the browser when opening the report.
Sanitization: There is no evidence of HTML escaping or sanitization of codebase strings before they are written to the architecture-review-*.html file.
[DYNAMIC_EXECUTION]: The HTML scaffold in HTML-REPORT.md explicitly configures the Mermaid.js library with securityLevel: "loose". This setting allows the execution of scripts within diagram definitions. If an attacker places malicious content in the codebase that the agent subsequently renders into a diagram, it could result in Cross-Site Scripting (XSS) within the user's browser context.
[EXTERNAL_DOWNLOADS]: The generated report fetches the Tailwind CSS and Mermaid.js libraries from external content delivery networks (CDNs), specifically cdn.tailwindcss.com and cdn.jsdelivr.net. These are well-known technology services used for legitimate reporting purposes.

improve-codebase-architecture