The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted data from the local codebase and incorporates it into a generated HTML report without explicit sanitization.
Ingestion points: Reads arbitrary source code files, CONTEXT.md, and architecture decision records (ADRs) from docs/adr/ within the scanned repository. Also processes git log output.
Boundary markers: None. The instructions do not specify any delimiters or safety markers to isolate codebase content within the HTML report template.
Capability inventory: Executes shell commands (git, xdg-open, open, start) and writes files to the system's temporary directory.
Sanitization: The skill lacks sanitization logic. Specifically, the HTML-REPORT.md template initializes the Mermaid diagramming library with securityLevel: "loose", which enables the execution of scripts embedded within diagram definitions if those definitions are derived from malicious code comments or identifiers found during the scan.
[COMMAND_EXECUTION]: The skill uses shell commands to perform analysis and interact with the operating system.
Evidence: Executes git log --oneline to identify codebase hot spots and uses platform-specific commands (xdg-open, open, start) to automatically launch the generated architectural review in the user's browser.
[EXTERNAL_DOWNLOADS]: The generated HTML report fetches resources from external services to render visuals.
Evidence: The report template in HTML-REPORT.md loads the Tailwind CSS framework from cdn.tailwindcss.com and the Mermaid.js library from cdn.jsdelivr.net at runtime.

improve-codebase-architecture