Source URL: https://simonwillison.net/2025/Mar/13/tools-colophon/
Source: Simon Willison’s Weblog
Title: My tools colophon now has AI-generated descriptions
Feedly Summary: My tools colophon now has AI-generated descriptions
The /colophon page on my tools site lists all 78 of my tools along with their commit histories, including links to prompting transcripts. I wrote about how I built that. the other day.
It now also includes a description of each tool, generated using Claude 3.7 Sonnet.
This is the single largest chunk of AI-generated text I’ve ever published on the indexable web. I share lots of LLM transcripts but I usually put them in a secret Gist such that they’ll be kept out of search engines.
On the one hand this feels slop adjacent – but I was never going to write documentation for all 78 of those little experiments, and yet find myself forgetting details of how they worked. Just one example: my CSV marker map tool which I built last year for knocking out an ad-hoc Bigfoot sightings map and then promptly forgot what it could do.
Here’s what I got when I ran:
cat csv-marker-map.html | llm -m claude-3.7-sonnet \
–system “Write a paragraph of documentation for this page as markdown. Do not include a heading. Do not use words like just or simply."
This interactive map displays markers based on data from a CSV file. Load data by adding ?csv=URL to the page URL, where the CSV must include latitude and longitude columns. You can customize the map with several URL parameters: center (coordinates in format lat,lng), zoom (zoom level), q (location to search for), marker (add individual markers at coordinates), and color (change the default marker color). The map automatically updates the URL as you navigate, preserving your current view. Markers from CSV files appear as small colored circles on the map, while markers specified directly in the URL use the default Leaflet marker style.
That’s genuinely useful. I got Claude to write me a write_docs.py script to loop through all 78 files and generate filename.docs.md descriptions that live in the repo, then updated build_colophon.py to include those descriptions if they exist.
It cost about $1.08 in Claude credits to document all 78 tools, because some of those HTML files were quite long so the input tokens added up.
Each of the documentation files includes a comment with the most recent commit hash of the file that was used to generate the document, like this:
The tool can use this to spot if a file has been updated – if so, the documentation will be regenerated.
Add this all together and now I can drop new HTML+JavaScript tools into my simonw/tools repo and, moments later, they’ll be published on tools.simonwillison.net with auto-generated descriptions added to my colophon. I think that’s pretty neat!
Tags: slop, claude, tools, projects, generative-ai, ai, llms, ai-assisted-programming
AI Summary and Description: Yes
Summary: The text discusses the implementation of AI-generated tool descriptions for a personal tool collection, utilizing an LLM (Claude 3.7 Sonnet) to create documentation for 78 tools. This innovative approach demonstrates how AI can enhance productivity and assist in documentation, providing valuable insights for professionals in AI and software development.
Detailed Description: The content highlights a personal project where the author employs an AI language model to automate the creation of documentation for various software tools. Below are the major points discussed:
– **AI-Generated Documentation**: The author uses Claude 3.7 Sonnet to generate descriptions for 78 tools on their website’s colophon page, marking a significant scale of AI-generated text on the indexable web.
– **Previous Practices**: While the author regularly shared LLM transcripts privately, this shift to public documentation allows others to access a comprehensive overview of their toolset.
– **Tool Example**: An illustrative example provided is the “CSV marker map” tool, which can create an interactive map using data from a CSV file. The generated documentation outlines its functionality and customization options clearly.
– **Efficiency and Cost**: The cost-effective nature of utilizing AI for documentation is emphasized, with the author noting that it only cost about $1.08 in AI credits to document all tools.
– **Version Control**: Each documentation file includes a comment with the commit hash of the source file used for generation, establishing a link between documentation and version control to ensure updates are reflected in the text.
This initiative showcases how AI can spearhead enhancements in software documentation and workflows, which can be particularly beneficial for developers, data scientists, and technology professionals experiencing the challenges of maintaining documentation for numerous tools or libraries. The integration of LLM capabilities highlights the productive relationship between AI and software development practices.
Key insights for professionals in the relevant fields include:
– Leveraging AI can dramatically streamline documentation processes.
– Real-time updates between documentation and source code can improve accuracy and maintainability.
– Cost-effective AI tools can enhance operational efficiency without significant resource investments.
Overall, this example serves as a practical application of AI in the realm of software security and documentation, reflecting broader trends in automation and AI assistance in technology workflows.