With VS Code

Arvo's own window is a separate application, and one day it may replace this workflow. Until then the engine sits behind a terminal and a folder of plain files, which is exactly what an editor is good at.

There is no Arvo extension for VS Code. You do not need one — the project is JSON files and the engine is a command.

The setup

Open the project folder as the workspace:

code ~/arvo

Everything you edit is a file in it: rules/, rulesets/, universes/, option-quotes/underlyings.json. Everything the engine produces is a file too: evidence/, sessions/, reviews/. Editing and reading are ordinary editor work.

flowchart LR
    subgraph VS ["VS Code workspace: ~/arvo"]
        ED["Editor<br/><small>rules/ · rulesets/ · universes/</small>"]
        TERM["Terminal<br/><small>arvo-engine …</small>"]
        OUT["evidence/ · sessions/ · reviews/"]
    end
    E["arvo-engine<br/><small>serving ~/arvo</small>"]

    ED -->|you write| E
    TERM -->|verbs| E
    E -->|writes| OUT
    OUT -->|you read| ED

Keep the engine in a terminal

Run it in a dedicated VS Code terminal so you can see what it is doing:

arvo-engine ~/arvo

Leave it. Use a second terminal for verbs. The engine outlives the editor — if you close VS Code with a session running, the session keeps trading, which is the intended behaviour and worth being deliberate about.

Schema help while editing

The project's JSON files have a shape, and the engine refuses a bad one with a reason. To get completion and errors while typing instead, point VS Code at the shape with a settings.json in the workspace:

.vscode/settings.json

{
  "files.associations": {
    "**/rules/*.json": "jsonc",
    "**/rulesets/*.json": "jsonc",
    "**/universes/*.json": "jsonc"
  },
  "files.exclude": {
    "**/engine.json": true,
    "**/control.json": true
  }
}

Hide control.json, and never commit it

It holds the token that can flatten your account. Excluding it from the explorer keeps it out of screenshots and out of the way; a .gitignore entry keeps it out of history.

.gitignore

engine.json
control.json
data/

data/ is excluded because it is large and refetchable. Do not exclude evidence/ or option-quotes/ — findings are your research history, and a recorded option chain cannot be fetched again after the day it was recorded.

Tasks for the verbs you run most

.vscode/tasks.json

{
  "version": "2.0.0",
  "tasks": [
    {
      "label": "Arvo: serve",
      "type": "shell",
      "command": "arvo-engine ${workspaceFolder}",
      "isBackground": true,
      "problemMatcher": []
    },
    {
      "label": "Arvo: leaderboard",
      "type": "shell",
      "command": "arvo-engine ${workspaceFolder} rank",
      "problemMatcher": []
    },
    {
      "label": "Arvo: review today",
      "type": "shell",
      "command": "arvo-engine ${workspaceFolder} review",
      "problemMatcher": []
    },
    {
      "label": "Arvo: sessions",
      "type": "shell",
      "command": "arvo-engine ${workspaceFolder} session list",
      "problemMatcher": []
    }
  ]
}

Deliberately absent: a task that starts a session or halts one. Both are decisions, and a decision does not belong behind a key you might hit by accident.

Claude Code in the same window

Claude Code's VS Code extension and the MCP server compose without any extra work — the agent gets Arvo's research tools in the editor you are already in.

claude mcp add arvo -- ~/.local/arvo/arvo-mcp-server ~/arvo --agent claude

The agent reads and writes the same files you have open, so a ruleset it writes appears in the explorer and a finding it produces appears under evidence/. It still cannot trade. With Claude Code →

Reading a finding

A finding is one JSON file, written to be read:

code ~/arvo/evidence/*.json

Read the advice and the verdict before any number — that ordering is the whole design. A Supported verdict with advice saying the average is carried by one instrument is not the result it looks like at a glance. Verdicts and advice →