improve-codebase-architecture

Pass

Audited by Gen Agent Trust Hub on Jul 19, 2026

Risk Level: SAFECOMMAND_EXECUTIONEXTERNAL_DOWNLOADSPROMPT_INJECTION
Full Analysis
  • [COMMAND_EXECUTION]: The skill uses operating system commands such as xdg-open, open, and start to automatically launch the generated architectural review in the user's web browser.\n- [EXTERNAL_DOWNLOADS]: The generated report fetches resources from established and well-known content delivery networks, including cdn.tailwindcss.com for styling and cdn.jsdelivr.net for the Mermaid diagramming library.\n- [PROMPT_INJECTION]: The skill processes untrusted repository data (e.g., CONTEXT.md and Architectural Decision Records) and embeds it into an HTML report. Because the Mermaid visualization is configured with securityLevel: 'loose' and no explicit sanitization is performed, this creates an indirect prompt injection surface where malicious repository content could potentially trigger cross-site scripting (XSS) in the user's browser.\n
  • Ingestion points: Ingests project-specific documentation from CONTEXT.md, docs/adr/, and codebase analysis results.\n
  • Boundary markers: No delimiters or instructions are used to separate untrusted content from the report template.\n
  • Capability inventory: Includes file system writes, system command execution, and subagent orchestration via the Agent tool.\n
  • Sanitization: The skill lacks logic to escape or filter content ingested from the codebase before including it in the report scaffold.
Audit Metadata
Risk Level
SAFE
Analyzed
Jul 19, 2026, 09:42 PM
Security Audit — agent-trust-hub — improve-codebase-architecture