The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted external data which creates an injection surface.
Ingestion points: Reads local codebase files, architectural decision records (ADRs) in docs/adr/, and git commit history via git log (SKILL.md).
Boundary markers: There are no explicit instructions or delimiters defined to prevent the agent from following instructions embedded within the analyzed codebase or commit messages.
Capability inventory: The skill can write HTML reports to the system's temporary directory, modify or create the GLOSSARY.md file, and execute shell commands to open files (SKILL.md).
Sanitization: No sanitization or escaping of the codebase content is specified before it is included in the generated HTML reports or the GLOSSARY.md file.
[COMMAND_EXECUTION]: The skill uses shell commands for legitimate development workflows.
Evidence: Executes git log --oneline to identify codebase hot spots (SKILL.md).
Evidence: Invokes platform-specific commands xdg-open, open, or start to display the generated architectural report to the user (SKILL.md).
[EXTERNAL_DOWNLOADS]: The generated HTML reports reference well-known external libraries for styling and visualization.
Evidence: Fetches the Tailwind CSS framework from https://cdn.tailwindcss.com (HTML-REPORT.md).
Evidence: Fetches the Mermaid diagramming library from https://cdn.jsdelivr.net/npm/mermaid@11/dist/mermaid.esm.min.mjs (HTML-REPORT.md).
[DYNAMIC_EXECUTION]: The report configuration increases the potential impact of cross-site scripting (XSS) in the generated report.
Evidence: The HTML scaffold explicitly sets Mermaid's securityLevel to "loose" (HTML-REPORT.md). This allows the rendering of arbitrary HTML tags within diagram nodes, which could lead to code execution in the user's browser if the analyzed codebase contains malicious diagram definitions.

improve-codebase-architecture