Open Source Work
I build open tools that help websites and institutions get ready for the agentic web, the internet as AI assistants and autonomous agents experience it.
My policy writing argued that tool using AI is a new category of actor on the web. This work is the practical side of that argument, published in the open at github.com/asish-singh.
agent-readiness-auditor
A command line tool that scans any website and scores how well it works with, and defends against, AI agents. It crawls the site and runs nine automated checks covering hidden prompt injection text, the presence of llms.txt, the robots.txt stance on AI crawlers, machine readable structured data, accountability links, sitemaps, content structured for AI answers, accidental index blocking, and whether an agent endpoint is advertised. Each audit returns a score out of 100, a letter grade, and a concrete suggested fix for every finding.
The auditor also ships as an MCP server, which means AI assistants such as Claude can run audits directly as a native tool. This very website was improved using its own findings.
ai-search-playbook
A public playbook on making websites visible to AI assistants and cited in AI answers. It explains the field in plain language, publishes my full audit methodology, includes a complete sample client report for a fictional company, and presents The State of the Agentic Web, a 2026 study of how prepared websites are for AI agents.
agentic-web-governance-pack
Governance artifacts for tool using AI agents, written for policy and procurement audiences. The pack covers risk assessment, transparency obligations, procurement language, and competition considerations. It connects my published position papers on AI policy with concrete documents that institutions can adapt and use.
freeyoutubetranscribe
Paste a YouTube link, read the video. A transcription tool that is free forever, live on the web, with a public write up of how it was designed and built.
Questions people ask
What is the agentic web?
The agentic web is the version of the internet used by AI assistants and autonomous agents rather than people with browsers. Agents read pages as data, follow machine readable signals like llms.txt and structured markup, and act on behalf of users. Websites that publish those signals get found, cited, and used by AI systems. Websites that do not become invisible to them.
What does the agent readiness auditor check?
Nine things. Hidden prompt injection text, the presence of llms.txt, the robots.txt stance on AI crawlers, structured data, accountability links, a sitemap, content structured for AI answers, accidental index blocking, and whether an agent endpoint such as an MCP server is advertised. It returns a score out of 100 with a letter grade and a suggested fix for every finding.
How can a website become visible to AI assistants?
Publish an llms.txt file listing key pages, declare a stance on AI crawlers in robots.txt, add schema.org structured data, provide a sitemap, and write content that answers questions directly. The full methodology is published openly in the ai-search-playbook repository. This website practices what it documents, and it scores 100 out of 100 on the auditor.