The Agent Skills Directory

[COMMAND_EXECUTION]: Executes shell commands including git log to analyze commit history and OS-specific commands (open, xdg-open, start) to display the generated HTML report.
[EXTERNAL_DOWNLOADS]: The generated report template fetches configuration and script assets from well-known technology services, including Tailwind CSS from cdn.tailwindcss.com and Mermaid.js from cdn.jsdelivr.net.
[PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection as it ingests untrusted data from the codebase to inform its analysis and report generation.
Ingestion points: Source code files, git commit history (git log), project metadata (CONTEXT.md), and Architecture Decision Records (docs/adr/).
Boundary markers: No explicit delimiters or boundary markers are defined to isolate untrusted codebase content from the agent's internal instructions.
Capability inventory: File system read/write access, git command execution, invocation of sub-agents for exploration, and the ability to open local files in the user's browser.
Sanitization: There is no evidence of content sanitization or validation for the data extracted from the codebase before it is used in the report or as input for further agent reasoning.
[COMMAND_EXECUTION]: The Mermaid.js configuration in the generated report template uses securityLevel: "loose", which allows the rendering of HTML within diagrams. This could potentially allow for cross-site scripting (XSS) in the local report if malicious strings from the codebase are included in the diagram definitions.

improve-codebase-architecture