ferdinandobons/brand-docs

BrandDocs is a set of agent skills for Claude Code and compatible AI coding assistants that solve a specific problem: when you ask an AI to generate a Word document, PowerPoint deck, or Excel workbook, it tends to invent its own formatting and only loosely imitate any template you show it. BrandDocs is built to prevent that drift by treating the template as a strict set of rules the AI must follow, not a loose inspiration. The workflow has two stages. First, you give it one of your company's existing Office files: a branded .docx.pptx, or .xlsx. The extraction step reads the file's internal structure, pulls out its theme colors, fonts, named styles, layouts, logos, and document skeleton (meaning the required order of sections like cover, table of contents, and body), and saves all of that into a portable folder called a brand kit. The original file is preserved as a shell. Second, when you want a new document, the generation step takes your content, maps each piece of it to the appropriate brand artifact from the profile, fills the shell in the correct structural order, and saves a new file. Because generation always opens from the real template shell and uses only the styles and colors defined there, the README describes it as "off-brand output impossible by construction." A verification step runs checks after every generation: it confirms that every style reference in the output actually exists in the template, that no placeholder or leftover template text survived, that tables are intact, and that Excel formulas were not broken when cells were filled. When LibreOffice is available, it can also render the output visually and flag layout problems. If rendering tools are not present, the tool reports that explicitly rather than skipping the check silently. All three Office formats use the same underlying engine and profile schema. The project is built on the standard Python libraries for reading and writing Office files (python-docx, python-pptx, openpyxl) with no external services or cloud dependencies. It is MIT licensed and marked alpha status at time of writing. The full README is longer than what was shown.