The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted data from the local project environment to generate its reports.
Ingestion points: Codebase files, git commit history (git log), and project-specific glossary files.
Boundary markers: The instructions do not define specific delimiters or guardrails to prevent the agent from interpreting instructions embedded within the codebase being scanned.
Capability inventory: The skill can spawn sub-agents, write files to the system's temporary directory, and execute shell commands to open those files.
Sanitization: No sanitization logic is specified for data extracted from the codebase before it is inserted into the HTML report scaffold.
[DYNAMIC_EXECUTION]: The HTML report template in HTML-REPORT.md explicitly sets the Mermaid.js securityLevel to "loose". This configuration allows for script execution and interactivity within diagrams. Because the report is populated with strings directly from the codebase (e.g., file names, problem descriptions), this creates a Cross-Site Scripting (XSS) vector if a file or commit message contains a malicious payload.
[COMMAND_EXECUTION]: The skill uses platform-specific commands (xdg-open, open, or start) to automatically launch the generated HTML report from the temporary directory in the user's default web browser.
[EXTERNAL_DOWNLOADS]: The generated report fetches dependencies from well-known services:
Loads Tailwind CSS for styling from https://cdn.tailwindcss.com.
Loads the Mermaid.js library for diagram rendering from https://cdn.jsdelivr.net/npm/mermaid.

improve-codebase-architecture