gs-improve-codebase-architecture
Pass
Audited by Gen Agent Trust Hub on Nov 20, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTIONCOMMAND_EXECUTIONEXTERNAL_DOWNLOADS
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill processes untrusted data from the local codebase, including source files, documentation, and git history, which can influence its analysis and the generated report content.
- Ingestion points: Reads project source code,
GLOSSARY.md, Architecture Decision Records indocs/adr/, and git history viagit log --oneline. - Boundary markers: Absent. The instructions do not define delimiters or specific constraints to ignore instructions embedded within the codebase files.
- Capability inventory: The skill can write files to the operating system's temporary directory and execute shell commands to open those files in the default browser.
- Sanitization: Absent. The generated HTML report reflects content from the codebase into diagrams and text. Furthermore, the Mermaid diagram configuration uses
securityLevel: "loose", which allows HTML tags in labels, potentially increasing the risk of cross-site scripting (XSS) if malicious content is pulled from the codebase. - [COMMAND_EXECUTION]: The skill executes shell commands to launch the generated architectural report for the user.
- Evidence: Uses
xdg-open(Linux),open(macOS), orstart(Windows) to open the temporary HTML file. - [EXTERNAL_DOWNLOADS]: The generated HTML report fetches external libraries from content delivery networks at runtime when opened in a browser.
- Evidence: The report template in
HTML-REPORT.mdincludes script tags forhttps://cdn.tailwindcss.comandhttps://cdn.jsdelivr.net/npm/mermaid@11/dist/mermaid.esm.min.mjs. These are well-known services used for styling and diagram rendering.
Audit Metadata