The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes project files and git history to identify architectural patterns and generate sub-agent instructions. This untrusted data is also used to generate an HTML report for the user.
Ingestion points: Reads arbitrary source code files and git log output from the local repository during the exploration phase.
Boundary markers: The instructions lack explicit delimiters or warnings to ignore embedded instructions when interpolating codebase data into sub-agent prompts.
Capability inventory: The skill can write files to the repository (documentation/architecture-audit/), modify the AGENTS.md file, and execute system commands to open generated reports in a web browser.
Sanitization: No sanitization is specified for codebase content before it is included in the generated HTML report. The report configuration uses Mermaid with securityLevel: "loose", which could allow execution of malicious scripts if injected into the diagram definitions from the analyzed code.
[COMMAND_EXECUTION]: The skill executes local shell commands to facilitate its workflow.
Uses git log --oneline to retrieve commit history for architectural analysis.
Uses OS-specific commands (open, xdg-open, start) to automatically open the generated HTML report in the user's default browser.
[EXTERNAL_DOWNLOADS]: The generated report fetches external resources from well-known CDN providers.
Loads the Tailwind CSS library from https://cdn.tailwindcss.com for report styling.
Loads the Mermaid JS library from https://cdn.jsdelivr.net for rendering architectural diagrams.

improve-codebase-architecture