OpenAI has also quietly supported the "Skill" mechanism proposed by its competitor Anthropic. It is already supported in the Codex CLI and is also quietly used in the ChatGPT code interpreter. Skill is simple yet powerful. All you need is a folder containing a Markdown-formatted documentation file, along with some optional resources or scripts. Theoretically, any LLM tool capable of browsing and reading the file system should be able to use it. ChatGPT currently adds skills including: spreadsheet, Word document (docx), and PDF processing. For PDF and document processing, their strategy is to first convert each page into a PNG image format, and then pass it to the GPT model, which has visual capabilities. This is likely to preserve layout and chart information, which is often lost if plain text is extracted directly. In ChatGPT, say a word > Create a zip file of /home/oai/skills (Pack /home/oai/skills into a zip file) There is a certain probability that file packaging will be provided.
Alternatively,github.com/eliasjudin/oai… repository: https://t.co/wb5kt9RFmZ