Skip to content

Route C — Agent Skills

Install the DOME Agent Skill into your agent of choice, point it at a DOI, and review what comes back.

Overview

Questions:

  • How do I install the DOME Agent Skill?
  • Does it only work with Claude, or with any agent?
  • What do I need on my machine before I start?
  • What does a run look like, and what do I check afterwards?

Learning Objectives

By the end of this chapter, you will be able to:

  1. Install the skill by whichever route fits your setup
  2. Run a DOME assessment from a DOI and read the resulting entry plus summary
  3. Interpret the source-provenance tags on each field
  4. Know what the skill does not yet do

Time: 35 minutes (hands-on exercise)

Prerequisite: Chapter 5 · Python 3.10+ · a terminal

Prototype — v0.1.0

Not yet benchmarked or validated. Frontier models give promising results and more capability than DOME Copilot, but read and correct every field before you use the output.

The repository

Everything lives in one public repository:

https://github.com/gavinf97/dome-agent-skill — CC BY 4.0, v0.1.0

dome-agent-skill/
├── SKILL.md                  # the 8-phase instructions the agent follows
├── references/               # DOME schemas v1.0.0 & v2.0.0, field guide,
│                             #   entry template & example, OSAI guidelines
├── scripts/                  # six standalone Python CLIs
│   ├── resolve_publication.py
│   ├── fetch_fulltext.py
│   ├── extract_pdf_text.py
│   ├── inspect_external_link.py
│   ├── suggest_osai_components.py
│   ├── submit_registry.py
│   └── requirements.txt
└── .claude-plugin/           # plugin + marketplace manifests

Not just for Claude

SKILL.md is ordinary Markdown: a set of instructions any capable agent can read and follow. The helper scripts return their results in JSON format, which any agent can work with. Claude is simply the quickest to set up.

Install

Open Claude Download the skill (ZIP)

In Claude Code — two commands in a session:

/plugin marketplace add gavinf97/dome-agent-skill
/plugin install dome-agent-skill@dome-marketplace

These are Claude Code commands; slash commands do not run in the browser chat.

In Claude in the browser — upload the skill once, then use it in any chat:

  1. Download the skill with the button above. It is already packaged the way Claude expects — nothing to unzip, nothing to rename.
  2. In the message box, click + → Skills → Manage skills, and upload dome-agent-skill.zip. (Browse skills, in the same menu, is the directory of ready-made skills.)
  3. Uploading your own skill needs a paid plan — Pro, Max, Team or Enterprise.
  4. The skill now appears under + → Skills. Just ask for what you want, for example “Assess this paper for DOME compliance: <DOI>“ — it triggers on intent, with no special syntax.

An IDE (integrated development environment) is the editor you write code in, such as VS Code. The agent built into it — GitHub Copilot, Gemini Code Assist, ChatGPT/Codex — can run this too.

git clone https://github.com/gavinf97/dome-agent-skill.git
cd dome-agent-skill
pip install -r scripts/requirements.txt

Then point the agent at the instructions:

Follow SKILL.md in this repository to assess the paper at <DOI or URL> against the DOME recommendations. Use the scripts in scripts/ for the lookups and file handling. Show me the entry and the notes summary.

The agent needs two abilities: read files and run commands. That is all.

Setup

# Python 3.10 or newer
python3 --version

# Three dependencies only: requests, PyMuPDF, PyYAML
pip install -r scripts/requirements.txt

Optionally, set GITHUB_TOKEN to raise the GitHub API rate limit for the external-link check in Phase 6. It is not required.

No Docker, no services

There is deliberately nothing to stand up. Three Python packages and a terminal.

Running it

Once installed, just ask in natural language:

Assess this paper for DOME compliance: https://doi.org/10.1038/s41592-021-01205-4

Or, with more context supplied up front:

Generate a DOME entry for DOI 10.1093/gigascience/giae094. The code is at https://github.com/BioComputingUP/dome-registry. Use schema v2.0.0.

The agent will work through the eight phases, then stop and show you two things.

What you get back

1. The entry JSON — conforming to dome-schema v2.0.0 (or v1.0.0 if you asked for it), ready for the Registry.

2. A notes and compliance summary — not part of the submitted JSON, containing:

  • Provenance breakdown — which fields came from paper, supplementary:<file>, or external:<url>. Read this first
  • Remaining gaps — fields left null, with a one-line reason for each
  • Improvement suggestions — concrete OSAI ecosystem components that would close each significant gap, cited by name, URL and OSAI recommendation code

external: is a finding, not a pass

A field the agent could only fill from a linked GitHub repository is tagged external:<url> and flagged as not disclosed in the paper itself. That is a DOME gap, correctly identified — the recommendations assess what the publication discloses. Do not quietly promote those fields to paper.

Submission

Submitting straight to the DOME Registry is not enabled. Review the output and take it forward yourself. A connection may follow once the approach is validated, but that is not confirmed.

When to choose this route

Good fit: one paper you care about, likely your own; you want DOME and OSAI recommendations together; supplementary material matters; you already work with an agentic assistant.

Poor fit: annotating at corpus scale (cost), no Python available, or a context where sending the paper to a frontier model is not acceptable — use Route B or Route A.

Sources

Next: Further Resources

  1. Gavin Farrell. DOME agent skill. 2026. Licensed CC BY 4.0. URL: https://github.com/gavinf97/dome-agent-skill. ↩