The Agent Skills Directory

[COMMAND_EXECUTION]: The skill generates a dynamic HTML report (architecture-review-<timestamp>.html) in the system's temporary directory and then automatically executes a platform-specific command (open, start, or xdg-open) to launch the file in a web browser.
[COMMAND_EXECUTION]: In the generated HTML report, the Mermaid.js library is initialized with securityLevel: "loose". This configuration allows the execution of scripts or the triggering of events through diagrams. If the agent unknowingly incorporates malicious content from the codebase into the diagrams, it could lead to cross-site scripting (XSS) or other local execution risks when the user opens the report.
[PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection because it reads and processes external, untrusted data including git log output and the project's source code files.
Ingestion points: Reads git log, CONTEXT.md, and walks the codebase using the Explore agent tool.
Boundary markers: The instructions do not specify the use of delimiters or specific warnings to ignore instructions found within the analyzed codebase.
Capability inventory: The skill has the ability to read arbitrary project files, write to the filesystem, and execute shell commands.
Sanitization: There is no explicit sanitization step described to prevent code or commit comments from being interpreted as architectural instructions or being directly injected into the visual HTML report.

improve-codebase-architecture