The Agent Skills Directory

[COMMAND_EXECUTION]: The skill executes shell commands to inspect the project history via git log --oneline and to display the generated HTML report using system-specific commands such as open, xdg-open, or start.
[DATA_EXPOSURE]: The skill accesses the local filesystem to read source code, commit history, and architectural documentation (CONTEXT.md, docs/adr/) to identify architectural friction points.
[EXTERNAL_DOWNLOADS]: The generated HTML report fetches resources from well-known services, specifically the Tailwind CSS framework and the Mermaid.js library via their respective CDNs.
[DYNAMIC_EXECUTION]: The skill generates an HTML report that initializes the Mermaid.js library with securityLevel: "loose". While this enables advanced diagramming features, it creates a potential risk if malicious content from the analyzed codebase is included in a diagram's labels, as the "loose" setting allows for more flexible rendering that could be exploited.
[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted input from the codebase being analyzed, which serves as a surface for indirect prompt injection.
Ingestion points: Project source files, git log output, and local markdown documentation such as CONTEXT.md and ADRs.
Boundary markers: The instructions do not specify the use of clear delimiters or instructions to treat codebase content as untrusted data.
Capability inventory: The skill can spawn sub-agents, write files to the temporary directory, execute shell commands, and invoke other architectural design tools.
Sanitization: There is no explicit mention of sanitizing codebase content before it is processed by the agent or embedded in the generated HTML report.

improve-codebase-architecture