Agents don't see your page. They read your text.
An AI agent never sees the rendered site, the styling, or the buttons. It reads the text in the payload. OSCAR's job is to make that text excellent and honest, and to keep it in plain sight for people too.
What a human sees
Sustainable Funding for Open Science
Home of the Open Science Assistant, a free AI tool for HED, BIDS, EEGLAB, and NEMAR researchers.
What an agent reads · /llms.txt
# OpenScience Collective > Sustainable funding for open-science software, and > home of the Open Science Assistant (OSA). ## Start here - [Open Science Assistant](/osa/): what OSA is - [Try the demo](https://demo.osc.earth): use OSA now - [Docs](https://docs.osc.earth) - [Courses](https://courses.osc.earth) ## Domain assistants - [HED](https://demo.osc.earth/hed): event annotation - [BIDS](https://demo.osc.earth/bids): data structure - [EEGLAB](https://demo.osc.earth/eeglab): EEG analysis - [NEMAR](https://demo.osc.earth/nemar): dataset search
Same information, two audiences, nothing hidden. The For AI agents link in our footer points straight to our llms.txt, the very index an agent reads, so anyone can audit exactly what we tell agents. No invisible text, no comment-only instructions, no cloaking.
Curated for the tools open science runs on
A "tool" is not one thing, so OSCAR gives a recipe per archetype. Most projects combine two or three.
Websites & docs sites
Curated llms.txt, an AI-aware robots.txt, JSON-LD, and clean markdown mirrors.
Command-line tools
Machine-readable --help, an AGENTS.md, and an optional MCP server.
Libraries & toolboxes
Example-led docstrings, accurate package metadata, and a docs llms.txt.
Web apps & APIs
An OpenAPI spec, an MCP server for the core actions, and SoftwareApplication JSON-LD.
Data archives
schema.org Dataset, a machine-readable catalog, and clear query, metadata, and download access.
Research lab sites
JSON-LD for people, software, and papers, with persistent identifiers (ORCID, ROR, DOI).
Standards & specs
A machine-readable schema, a callable validator, and a versioned, citable specification.
Transparency by construction
Everything OSCAR recommends is public, human-readable, and linked from the site. If it is for agents, it is visible to people. We actively reject the alternative: hidden text, invisible instructions, or comment-only directives aimed at agents are a form of indirect prompt injection, ranked first in the OWASP Top 10 for Large Language Model Applications, and an open-science project has more to lose there than anyone.
The honest way to guide an agent is to do it in the open, in a file a person can click and read. That is what this page, and the footer link, are for.
Make your tool agent-ready
OSCAR is a public good. Clone the standard, follow the guide for your archetype, and self-audit against the checklist.