The Agent Skills Directory

[PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection because it ingests untrusted codebase data, such as file paths and ADR content, and incorporates it into a generated HTML report.
Ingestion points: Codebase files, CONTEXT.md, and ADRs accessed via the Explore subagent tool.
Boundary markers: Absent. The skill does not use delimiters or instructions to sanitize the untrusted content before rendering it in HTML.
Capability inventory: The skill possesses file read/write access and the ability to execute system commands to open files.
Sanitization: The report template in HTML-REPORT.md uses securityLevel: "loose" for Mermaid diagrams. This setting disables the library's internal sandbox, creating a potential vector for Cross-Site Scripting (XSS) if the codebase contains malicious diagram syntax.
[EXTERNAL_DOWNLOADS]: The generated HTML report fetches resources from well-known services, specifically Tailwind CSS from cdn.tailwindcss.com and the Mermaid library from cdn.jsdelivr.net. These are standard resources used for report layout and visualization.
[COMMAND_EXECUTION]: The skill executes git log to analyze project history and uses platform-specific commands (xdg-open, open, or start) to launch the generated report for the user. These operations are limited to standard development tools and the skill's own generated outputs.

improve-codebase-architecture