The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted content from the codebase, including source code, ADRs, and project documentation, creating a surface for embedded instructions.
Ingestion points: Reads local repository files, CONTEXT.md, and docs/adr/.
Boundary markers: No explicit markers or instructions to ignore embedded instructions are defined for the analysis phase.
Capability inventory: Reading and writing local files, executing git commands, and invoking system browsers.
Sanitization: No explicit sanitization of codebase data is performed before inclusion in the generated HTML report.
[EXTERNAL_DOWNLOADS]: The skill generates a report that fetches assets from well-known services.
Downloads the Tailwind CSS framework from cdn.tailwindcss.com.
Fetches the Mermaid.js library for diagram rendering from cdn.jsdelivr.net.
[COMMAND_EXECUTION]: The skill utilizes standard system tools to perform its functions.
Executes git log --oneline to identify codebase hot spots based on commit history.
Uses OS-specific commands such as open, xdg-open, or start to display the generated HTML report.
[DYNAMIC_EXECUTION]: The skill creates and renders content dynamically at runtime.
Generates a self-contained HTML report file in the system's temporary directory.
Configures the Mermaid diagramming tool with securityLevel: "loose", which permits the rendering of HTML tags within diagrams and could lead to script execution if malicious content from the codebase is included in diagram definitions.

improve-codebase-architecture