Skip to content
OstenHunPublic

About

Local research evidence CLI: connect paper claims to project decisions and export context for AI agents.

Topics

Resources

Contributing

Security policy

Stars

22 stars

Watchers

0 watching

Forks

Latest commit

 

History

72 Commits

Folders and files

NameName
Last commit message
Last commit date
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 

Repository files navigation

Danseo

Danseo — research evidence, project decisions.

See a paper graph

Danseo uses your AI agent to map how papers connect and where to read the evidence. The optional CLI saves and compares evidence across a research topic.

Nine-paper Transformer map with the focal paper's reading guidance, dates, and source locations

This preview connects Attention Is All You Need to earlier work on sequence modeling and to later papers including BERT, T5, and Vision Transformer. Solid green lines show source-checked connections. Dashed lines show suggestions or connections that need a source check. View the full-size screenshot.

License: MIT Node.js 22 or 24 LTS Development preview

Korean README · Documentation · Contribute · Open an issue

Install the skill

You need Node.js 24.x or 22.x LTS and a signed-in local AI agent that can read files and run shell commands. To start from a paper title, the agent also needs to be able to find papers online and open their sources. For papers you already have, give the agent access to the local PDFs.

The graph skill runs directly from its copied folder, with no npm install, build, or server.

1. Download the source

If you have Git, run these commands in a terminal:

git clone https://github.com/OstenHun/danseo.git
cd danseo
node --version

You can also use Code → Download ZIP on GitHub. Extract the ZIP, open a terminal in the folder containing README.md and skills/, and run node --version. You should see a version starting with v24. or v22..

2. Choose your agent and operating system

From the downloaded source folder, run the block for your operating system and enter the choice shown for your agent. The commands copy the complete skill folder to your home directory so Danseo is available across projects.

Windows · PowerShell
& {
    $danseoHost = Read-Host 'Choose codex, claude, or agy'
    $danseoLocations = @{
        codex  = '.agents/skills/danseo'
        claude = '.claude/skills/danseo'
        agy    = '.gemini/config/skills/danseo'
    }
    if (-not $danseoLocations.ContainsKey($danseoHost)) { throw 'Enter codex, claude, or agy.' }

    $danseoTarget = Join-Path $HOME $danseoLocations[$danseoHost]
    if (Test-Path -LiteralPath $danseoTarget) { throw "Danseo is already installed at $danseoTarget" }
    New-Item -ItemType Directory -Force -Path (Split-Path $danseoTarget -Parent) -ErrorAction Stop | Out-Null
    Copy-Item -LiteralPath './skills/danseo' -Destination $danseoTarget -Recurse -ErrorAction Stop
    Write-Output "Installed Danseo at $danseoTarget"
}
macOS or Linux · Bash or Zsh
{
  printf 'Choose codex, claude, or agy: '
  read -r danseohost
  case "$danseohost" in
    codex)  danseotarget="$HOME/.agents/skills/danseo" ;;
    claude) danseotarget="$HOME/.claude/skills/danseo" ;;
    agy)    danseotarget="$HOME/.gemini/config/skills/danseo" ;;
    *)      danseotarget='' ;;
  esac

  if [ -z "$danseotarget" ]; then
    printf '%s\n' 'Enter codex, claude, or agy.'
  elif [ -e "$danseotarget" ] || [ -L "$danseotarget" ]; then
    printf 'Danseo is already installed at %s\n' "$danseotarget"
  else
    mkdir -p "$(dirname "$danseotarget")" &&
      cp -R ./skills/danseo "$danseotarget" &&
      printf 'Installed Danseo at %s\n' "$danseotarget"
  fi
}

The script prints Installed Danseo at followed by the skill folder path. If you already have Danseo installed, check which copy your agent loads. To update, move that copy to a backup location outside the agent's skill folders and run the copy instructions again.

Generate a paper-relationship graph

3. Create and open your first graph

Start a new chat in a folder where your agent can save files. Try this request:

I want to read Attention Is All You Need. Find a few related papers from
before and after it. Create an interactive HTML graph in English and save it
as attention-paper-graph.html in this folder. Show the source location for
checked connections and label reading suggestions separately. Add a short
reason to read each paper. Use source records to verify first-publication
and selected-edition dates when available. Give me the path to the HTML file.

If your agent does not pick up Danseo, choose it in the Codex skill picker or start the message with $danseo. In Claude Code or Antigravity CLI, start the message with /danseo and add the prompt.

Open attention-paper-graph.html in a desktop browser. Select the focal paper for its reading guidance and any available page or section references. Select a source-checked connection to see its evidence. Which related papers appear depends on the sources your agent can access.

Paper nodes show the verified first-publication year when available. Paper details show the date and its source, along with the selected edition's date when verified.

When your agent verifies a venue and its rank, the graph shows a representative badge. Expand the paper's additional rankings to see other verified grades and their sources. The guide explains the ICORE, JCR, and Norwegian Register scales. JCR details include the subject category when available. The T5 badge in the preview comes from the China Computer Federation (CCF). Compare grades within the same ranking system.

If the agent can't access online sources, give it absolute paths to two or more local PDFs it can read. Ask for a graph using those files and have it report any connections it could not check. Review your agent's file permissions and provider settings before sharing private PDFs.

If Danseo doesn't appear, restart your agent and check that the installed folder contains SKILL.md, scripts/, references/, and assets/. You can also give the agent the absolute path to the installed SKILL.md and ask it to use that skill.

For more ways to start, see paper-reading orientation and the Danseo skill guide.

Quick start

Build and try the CLI

The CLI saves paper selections, source files or links, answers and notes, and evidence for comparisons. It runs independently of skill installation. Use Node.js 24.x or 22.x LTS with npm. From the downloaded source folder, run:

npm ci
npm run build
node ./dist/cli.js validate --workspace ./examples/minimal --json

The first command fetches the locked dependencies. After the build, the example should report ok: true, three records, two projects, and one expected draft warning. These checks cover file structure and references. The example uses synthetic cards with Korean body text. The CLI messages are in English.

For your own research, follow the CLI and agent guide to create a topic folder, collect sources, and open a local comparison. For reusable Paper and Application cards, see connect your own workspace.

To continue a search with the connected CLI in a new chat, give your agent the existing topic folder path and ask for another search round. The folder keeps your earlier rounds and selections.

Companion skills and output language

  • danseo-reading helps you ask questions about a paper and explore its concepts. Its view generator needs Node.js.
  • paper-to-project guides you through comparing a paper's conditions with your project. It needs no Node.js runtime.

To add a companion skill, use the copy instructions above with the companion's folder name in the source path and each destination path. Start a new chat and invoke it by that name.

Ask for the language you want in the prompt. Standalone graph and learning views have English and Korean controls. They preserve original paper titles and quotations. The CLI and its comparison viewer have English controls. See output-language rules.

Project structure

Danseo is a TypeScript/Node.js package with portable agent skills. A topic folder keeps the selected papers, collected sources, and notes for one research question. A separate workspace.json registers reusable Paper records and project-specific Application records. Several projects can cite the same evidence while making different decisions about it.

src/                     CLI, file-based core, comparison server and browser UI
schemas/                 Card, workspace, and relationship schemas
skills/                  Research graphs, paper learning, and project application
examples/minimal/        Synthetic workspace for validation and export
tests/                   CLI, source and package checks
docs/                    Usage guides, data formats and development references

Keep private papers and records outside the source repository. Exports can include full card text and absolute local paths, so review them before sharing.

Status and limitations

Danseo is a development preview. The CLI and file formats may change. Build the CLI from source. The read-only comparison viewer runs locally in a desktop browser. Stop its server with Ctrl+C. Your agent handles web search and source access under its own permissions.

Source checks cover the displayed evidence. Results depend on which sources the agent can read, and scientific reproduction is a separate step. See runtime support, current scope, and distribution.

Documentation

Goal Start here
Generate a paper graph First graph · Skill guide
Find, collect, analyze, and compare papers CLI and agent guide
Look up file formats and workflows Documentation index
Use the logos and banners Brand assets

Contributing and license

Bug reproductions, clearer documentation, synthetic examples, and reports from other environments are welcome. Read the first contribution guide or browse good first issues. Use the security policy to report vulnerabilities privately.

Code, documentation, skills, and synthetic examples use the MIT License. Papers, datasets, and user-supplied material retain their own terms. See third-party notices.

About

Local research evidence CLI: connect paper claims to project decisions and export context for AI agents.

Topics

Resources

Contributing

Security policy

Stars

22 stars

Watchers

0 watching

Forks

Releases

Packages

Contributors

Languages