Skip to content

OpenScience-Collective/oscar

Repository files navigation

OSCAR

OSCAR

Open Science Collective Agent Readiness
Make your open-science tools legible to AI agents, transparently.


OSCAR is the best practices, rules, and tools that make an open-science project ready for AI agents. It helps any maintainer answer two questions:

  1. How do agents find out my tool exists, and what it does?
  2. How do agents use my tool to the fullest once they have found it?

It is a resource of the OpenScience Collective, a sibling to the Open Science Assistant (OSA). Where OSA is an assistant that uses open-science tools, OSCAR is the standard that makes those tools usable by any assistant.

The one rule: transparency by construction

Everything OSCAR asks you to publish is public, human-readable, and linked from your site. Anything an agent is told, a person can open and audit.

OSCAR does not use, and actively warns against, hidden text, invisible instructions, or comment-only directives aimed at agents. Those techniques are a form of indirect prompt injection; security teams catalog them as an attack, and they erode the trust open science runs on. See docs/principles.md.

What agents actually see

An agent never sees your rendered page, your styling, or your buttons. It reads the text in the payload. OSCAR's job is to make that text excellent and honest.

What a human sees                 What an agent reads (/llms.txt)
-----------------                 ------------------------------
[ rendered page, CSS, buttons ]   # Your Tool
                                  > One-line description of what it is.
  "Try the demo"  ->  button      ## Start here
                                  - [Docs](https://...): ...
                                  - [Demo](https://...): try it now

Not a fad: FAIR made practical

Machine-actionability is not new. The FAIR principles (2016) already called for data usable by an "autonomously-acting, computational data explorer." OSCAR operationalizes FAIR's Interoperable and Reusable principles for a new such explorer, the AI agent, and extends them from datasets to every kind of tool. It also closes a gap: parts of the scholarly web solved structured metadata years ago but now regress into client-rendered app shells an agent cannot read, while the newer agent conventions have barely reached open science at all. See docs/grounding.md.

The archetypes

A "tool" is not one thing, so OSCAR gives a recipe per archetype. Most projects are a combination of two or three.

Archetype Example Primary techniques
Website / docs site eeglab.org, nemar.org llms.txt, AGENTS.md, AI-friendly robots.txt, JSON-LD, markdown mirrors
Command-line tool nemar-cli machine-readable --help, AGENTS.md, examples cookbook, optional MCP server
Library / toolbox EEGLAB (MATLAB) docstrings, typed signatures, docs llms.txt, MCP server, worked notebooks
Web app / API HEDit, OSA OpenAPI spec, MCP server, JSON-LD SoftwareApplication, clear auth and limits
Data archive NEMAR schema.org Dataset, machine-readable catalog, query + metadata + bulk download
Research lab / project site a lab or project website JSON-LD Organization/Person/SoftwareSourceCode, persistent IDs (ORCID, ROR, DOI)
Standard / specification BIDS, HED machine-readable schema, a callable validator, versioned citable spec

OSCAR curates instructions for the kinds of project the OpenScience Collective services: data archives, tools and toolsets, research lab and project sites, and standards. The set is meant to be extensible; new archetypes and worked examples are added as the landscape grows. Full guides live in docs/archetypes/.

Where each file goes

Two files are site-wide and live at the origin root; everything else is per-resource.

Artifact Location Scope
robots.txt site root, /robots.txt one per origin, never per-page
llms.txt site root, /llms.txt one per site; links out to per-section pages
AGENTS.md repo root, optionally site root one per repo; may nest per subdirectory, closest wins
JSON-LD / schema.org each page's <head>, server-rendered per page
Markdown mirror beside each page, /x to /x.md per page
Signposting Link HTTP header or <link> per resource per resource

You never fork llms.txt or robots.txt per page. Per-page detail is the JSON-LD's job: a homepage carries Organization, a dataset page Dataset, a tool page SoftwareApplication. The one root llms.txt gains depth by linking to those pages, not by cloning itself.

Layout

oscar/
  docs/            The doctrine: principles + one guide per archetype + a checklist
  templates/       Copy-paste starter files (llms.txt, AGENTS.md, robots.txt, JSON-LD)
  examples/        Worked case studies (HEDit, EEGLAB, NEMAR)
  brand/           Logo and favicon

Quick start

  1. Read docs/principles.md, the non-negotiables.
  2. Find your archetype(s) in docs/archetypes/.
  3. Copy the matching starter files from templates/ and fill them in.
  4. Self-audit against docs/checklist.md.
  5. Link the files from your site footer so they are auditable.

Status

Early and public. The principles and the archetype framework are stable; the per-archetype guides and worked examples are being written in the open. See .context/plan.md for the roadmap.

License

Documentation and templates: Creative Commons Attribution 4.0 (CC-BY-4.0). Code (generators, scripts): MIT. See LICENSE.

About

Open Science Collective Agent Readiness: make your open-science tools legible to AI agents, transparently.

Resources

License

Code of conduct

Contributing

Security policy

Stars

0 stars

Watchers

0 watching

Forks

Releases

No releases published

Packages

 
 
 

Contributors