AI answers are easy to get but expensive to read — a CLI that filters them down

An AI assistant provides an answer to a question, but the process of reading the lengthy response can be costly. The author found it frustrating to ask for a simple solution, like removing duplicates from a CSV in Python, and receive hundreds of characters detailing BOMs, encodings, and pandas design philosophy. The author then created a command-line interface (CLI) to filter the AI's response, making it more manageable.

The CLI takes a Markdown-formatted answer and produces three sections: a conclusion, minimal code, and verification instructions. It processes the input via stdin, allowing users to pipe the answer directly into it. The tool is available at https://github.com/sunnydachs/concise-md. The implementation is straightforward Markdown parsing, splitting the input into sections, detaching code fences from prose, and reorganizing the text.

The conclusion is determined automatically, based on specific markers like "tl;dr" or "the standard approach." Code blocks are preserved exactly as they appear in the original answer. For verification, the CLI looks for headings like "Verification" or "How to verify," or sentences containing the words "verify" or "check." If the answer is short and contains only one code block, the CLI passes the input through unchanged.

The tool is tested on twelve different commits across Python versions 3.11-3.13. The author acknowledges the limitations of the Markdown parser, which is based on regexes rather than a full CommonMark implementation, potentially causing odd results when parsing tables-heavy answers. Additionally, the tool cannot determine the correctness of the content, so any wrong answer will be shortened, but still incorrect.