The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted information from the codebase during analysis, creating a surface for indirect prompt injection.
Ingestion points: Reads local source files, commit history via git log, and architectural documentation (CONTEXT.md and docs/adr/) as specified in SKILL.md.
Boundary markers: The instructions do not define delimiters or specific warnings to prevent the agent from following instructions potentially embedded in the analyzed code.
Capability inventory: The agent can execute shell commands (git, xdg-open), read the filesystem, and write files to the system's temporary directory (SKILL.md).
Sanitization: There is no explicit requirement to sanitize or escape data extracted from the codebase before including it in the HTML report or Mermaid diagrams.
[DYNAMIC_EXECUTION]: The generated HTML report scaffold uses a configuration that permits script execution within the diagramming context.
Evidence: In HTML-REPORT.md, the Mermaid library is initialized with securityLevel: "loose".
Impact: This setting allows Mermaid to execute JavaScript or handle interactive elements within diagrams. If malicious diagram code is present in the analyzed codebase and included in the report, it could execute in the user's browser context upon opening the file.
[EXTERNAL_DOWNLOADS]: The architectural report fetches necessary assets and libraries from well-known services.
Evidence: The report scaffold in HTML-REPORT.md includes scripts from cdn.tailwindcss.com and cdn.jsdelivr.net (for Mermaid).
[COMMAND_EXECUTION]: The skill uses standard command-line tools to interact with the repository and the user's operating system.
Evidence: Executes git log --oneline to analyze hotspots and uses platform-specific commands like xdg-open, open, or start to display the final report.

improve-codebase-architecture