The Agent Skills Directory

[COMMAND_EXECUTION]: The skill uses shell commands to interact with the environment, including git log --oneline for history analysis and platform-specific commands (xdg-open, open, start) to launch the generated HTML report for the user.
[EXTERNAL_DOWNLOADS]: The generated report template in HTML-REPORT.md references external assets from cdn.tailwindcss.com and jsdelivr.net (Mermaid). While these are recognized services, they introduce external dependencies into the report's execution context.
[DYNAMIC_EXECUTION]: The Mermaid initialization in HTML-REPORT.md sets securityLevel: "loose". This configuration disables the default sandbox protections in Mermaid, potentially allowing malicious scripts embedded within the generated diagram data (sourced from the codebase) to execute in the browser when the report is opened.
[INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and process untrusted data from the local codebase.
Ingestion points: The agent reads the codebase files, git log output, GLOSSARY.md, and Architecture Decision Records (ADRs).
Boundary markers: The instructions do not define specific delimiters or escaping mechanisms to prevent instructions embedded in the codebase from influencing the agent's behavior.
Capability inventory: The skill can write files to the system temporary directory, modify GLOSSARY.md, create ADRs, and execute shell commands to open files.
Sanitization: There is no evidence of sanitization or filtering for data extracted from the codebase before it is used to generate reports or update project documentation.

improve-codebase-architecture