The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted data from the local codebase, including file contents, git history, and domain glossaries. A malicious repository could contain adversarial content designed to influence the agent's architectural analysis.
Ingestion points: Codebase files, git log output, CONTEXT.md, and Architecture Decision Records (ADRs).
Boundary markers: No specific delimiters or safety instructions are defined for the interpolation of codebase data into the agent's context.
Capability inventory: File system read/write (to temp directories), subprocess execution (git, open), and the ability to spawn sub-agents for exploration.
Sanitization: The instructions do not specify sanitization or escaping of project data before it is rendered into the HTML report or processed by the agent.
[DYNAMIC_EXECUTION]: The generated HTML report initializes the Mermaid diagramming library with securityLevel: "loose". This setting allows for expanded diagram features but reduces protection against potentially malicious scripts if adversarial codebase content (such as file names or labels) is included in the diagram structure.
[EXTERNAL_DOWNLOADS]: The generated architecture report fetches styling and diagramming logic from well-known content delivery networks at runtime.
Resources: Fetches Tailwind CSS from https://cdn.tailwindcss.com and Mermaid.js from https://cdn.jsdelivr.net.
[COMMAND_EXECUTION]: The skill automatically executes system commands to open the generated report in the user's default browser.
Commands: Utilizes xdg-open on Linux, open on macOS, and start on Windows with agent-generated file paths located in the system temporary directory.

improve-codebase-architecture