주고받는 이메일 데이터에 AI를 활용하는 것에 대한 이야기가 나왔다.

하지만, 이걸 위해 메일 서버를 별도로 구축하거나, 메일 클라이언트와 검색 기능 등을 별도로 코딩하기에는 아무리 AI Code Generation을 쓴다고 해도, 쓸만한 결과물이 나오기까지의 과정이 여간 쉬운 일이 아니다.

결국, 이메일과 관련된 모든 기능이 이미 있는 "MS Office"에 붙어서 바로 코딩할 수 있는 JS 프레임워크를 이용하기로 했다.

MS Outlook의 메일을 AI로 분석하는 실제 예시

// Analyze Microsoft Outlook data with ChatGPT
// Require: WelsonJS framework (https://github.com/gnh1201/welsonjs)
// Workflow: Microsoft Outlook -> OpenAI -> Get Response
var Office = require("lib/msoffice");
var LIE = require("lib/language-inference-engine");

function main(args) {
    var prompt_texts = [];

    var keyword = "example.com";
    var maxCount = 10;
    var previewLen = 160;

    console.log("Searching mails by sender OR recipient contains: '" + keyword + "'.");
    console.log("This test uses Restrict (Sender/To/CC/BCC) + Recipients verification.");
    console.log("Body preview length: " + previewLen);

    var outlook = new Office.Outlook();
    outlook.open();
    outlook.selectFolder(Office.Outlook.Folders.Inbox);

    var results = outlook.searchBySenderOrRecipientContains(keyword);
    console.log("Printing search results. (max " + maxCount + ")");

    results.forEach(function (m, i) {
        var body = String(m.getBody() || "");
        var preview = body.replace(/\r/g, "").replace(/\n+/g, " ").substr(0, previewLen);
        
        var text = "#" + String(i) +
            " | From: " + String(m.getSenderEmailAddress()) +
            " | To: " + String(m.mail.To || "") +
            " | Subject: " + String(m.getSubject()) +
            " | Received: " + String(m.getReceivedTime());

        console.log(text);
        console.log("  Body: " + preview);
        
        // Add an email data to the prompt text context
        prompt_texts.push(text);
        
        // The body to reduce token usage and avoid sending overly large/sensitive content.
        var bodyForPrompt = body;
        var maxBodyLengthForPrompt = 2000; // Keep the body snippet short
        if (bodyForPrompt.length > maxBodyLengthForPrompt) {
            bodyForPrompt = bodyForPrompt.substring(0, maxBodyLengthForPrompt) + "...";
        }
        prompt_texts.push("  Body: " + bodyForPrompt);
    }, maxCount);

    outlook.close();

    // build a AI prompt text
    var instruction_text = "This is an email exchange between the buyer and me, and I would appreciate it if you could help me write the most appropriate reply.";
    prompt_texts.push(instruction_text);

    // complete the prompt text
    var prompt_text_completed = prompt_texts.join("\r\n");

    //console.log(prompt_text_completed);  // print all prompt text

    // get a response from AI
    var response_text = LIE.create().setProvider("openai").inference(prompt_text_completed, 0).join(' ');

    console.log(response_text);
}

exports.main = main;

실행 방법

1. CLI 사용

모든 작성 및 저장을 마친 후, 다음 명령을 통해 실행한다. (outlook_ai.js 파일로 저장했을 때.

cscript app.js outlook_ai

2. GUI 사용

모든 작성 및 저장을 마친 후, WelsonJS Launcher 앱을 통해 실행한다.

실행하면 어떤 결과가 나오는가?

메일 내용에는 개인정보가 포함되어 있으므로 예시는 따로 첨부하지 않았다.

위 코드의 작업이 성공하면 메일 내용이 출력되면서, OpenAI 서버에서 분석을 마친 결과값을 얻어올 수 있다.

Consider Git's -C option:

git -C /path/to/repo checkout <TAB>

When you hit Tab, Git completes branch names from /path/to/repo, not your current directory. The completion is context-aware—it depends on the value of another option.

Most CLI parsers can't do this. They treat each option in isolation, so completion for --branch has no way of knowing the --repo value. You end up with two unpleasant choices: either show completions for all possible branches across all repositories (useless), or give up on completion entirely for these options.

Optique 0.10.0 introduces a dependency system that solves this problem while preserving full type safety.

Static dependencies with or()

Optique already handles certain kinds of dependent options via the or() combinator:

import { flag, object, option, or, string } from "@optique/core";

const outputOptions = or(
  object({
    json: flag("--json"),
    pretty: flag("--pretty"),
  }),
  object({
    csv: flag("--csv"),
    delimiter: option("--delimiter", string()),
  }),
);

TypeScript knows that if json is true, you'll have a pretty field, and if csv is true, you'll have a delimiter field. The parser enforces this at runtime, and shell completion will suggest --pretty only when --json is present.

This works well when the valid combinations are known at definition time. But it can't handle cases where valid values depend on runtime input—like branch names that vary by repository.

Runtime dependencies

Common scenarios include:

• A deployment CLI where --environment affects which services are available
• A database tool where --connection affects which tables can be completed
• A cloud CLI where --project affects which resources are shown

In each case, you can't know the valid values until you know what the user typed for the dependency option. Optique 0.10.0 introduces dependency() and derive() to handle exactly this.

The dependency system

The core idea is simple: mark one option as a dependency source, then create derived parsers that use its value.

import {
  choice,
  dependency,
  message,
  object,
  option,
  string,
} from "@optique/core";

function getRefsFromRepo(repoPath: string): string[] {
  // In real code, this would read from the Git repository
  return ["main", "develop", "feature/login"];
}

// Mark as a dependency source
const repoParser = dependency(string());

// Create a derived parser
const refParser = repoParser.derive({
  metavar: "REF",
  factory: (repoPath) => {
    const refs = getRefsFromRepo(repoPath);
    return choice(refs);
  },
  defaultValue: () => ".",
});

const parser = object({
  repo: option("--repo", repoParser, {
    description: message`Path to the repository`,
  }),
  ref: option("--ref", refParser, {
    description: message`Git reference`,
  }),
});

The factory function is where the dependency gets resolved. It receives the actual value the user provided for --repo and returns a parser that validates against refs from that specific repository.

Under the hood, Optique uses a three-phase parsing strategy:

1. Parse all options in a first pass, collecting dependency values
2. Call factory functions with the collected values to create concrete parsers
3. Re-parse derived options using those dynamically created parsers

This means both validation and completion work correctly—if the user has already typed --repo /some/path, the --ref completion will show refs from that path.

Repository-aware completion with @optique/git

The @optique/git package provides async value parsers that read from Git repositories. Combined with the dependency system, you can build CLIs with repository-aware completion:

import {
  command,
  dependency,
  message,
  object,
  option,
  string,
} from "@optique/core";
import { gitBranch } from "@optique/git";

const repoParser = dependency(string());

const branchParser = repoParser.deriveAsync({
  metavar: "BRANCH",
  factory: (repoPath) => gitBranch({ dir: repoPath }),
  defaultValue: () => ".",
});

const checkout = command(
  "checkout",
  object({
    repo: option("--repo", repoParser, {
      description: message`Path to the repository`,
    }),
    branch: option("--branch", branchParser, {
      description: message`Branch to checkout`,
    }),
  }),
);

Now when you type my-cli checkout --repo /path/to/project --branch <TAB>, the completion will show branches from /path/to/project. The defaultValue of "." means that if --repo isn't specified, it falls back to the current directory.

Multiple dependencies

Sometimes a parser needs values from multiple options. The deriveFrom() function handles this:

import {
  choice,
  dependency,
  deriveFrom,
  message,
  object,
  option,
} from "@optique/core";

function getAvailableServices(env: string, region: string): string[] {
  return [`${env}-api-${region}`, `${env}-web-${region}`];
}

const envParser = dependency(choice(["dev", "staging", "prod"] as const));
const regionParser = dependency(choice(["us-east", "eu-west"] as const));

const serviceParser = deriveFrom({
  dependencies: [envParser, regionParser] as const,
  metavar: "SERVICE",
  factory: (env, region) => {
    const services = getAvailableServices(env, region);
    return choice(services);
  },
  defaultValues: () => ["dev", "us-east"] as const,
});

const parser = object({
  env: option("--env", envParser, {
    description: message`Deployment environment`,
  }),
  region: option("--region", regionParser, {
    description: message`Cloud region`,
  }),
  service: option("--service", serviceParser, {
    description: message`Service to deploy`,
  }),
});

The factory receives values in the same order as the dependency array. If some dependencies aren't provided, Optique uses the defaultValues.

Async support

Real-world dependency resolution often involves I/O—reading from Git repositories, querying APIs, accessing databases. Optique provides async variants for these cases:

import { dependency, string } from "@optique/core";
import { gitBranch } from "@optique/git";

const repoParser = dependency(string());

const branchParser = repoParser.deriveAsync({
  metavar: "BRANCH",
  factory: (repoPath) => gitBranch({ dir: repoPath }),
  defaultValue: () => ".",
});

The @optique/git package uses isomorphic-git under the hood, so gitBranch(), gitTag(), and gitRef() all work in both Node.js and Deno.

There's also deriveSync() for when you need to be explicit about synchronous behavior, and deriveFromAsync() for multiple async dependencies.

Wrapping up

The dependency system lets you build CLIs where options are aware of each other—not just for validation, but for shell completion too. You get type safety throughout: TypeScript knows the relationship between your dependency sources and derived parsers, and invalid combinations are caught at compile time.

This is particularly useful for tools that interact with external systems where the set of valid values isn't known until runtime. Git repositories, cloud providers, databases, container registries—anywhere the completion choices depend on context the user has already provided.

This feature will be available in Optique 0.10.0. To try the pre-release:

deno add jsr:@optique/core@0.10.0-dev.311

Or with npm:

npm install @optique/core@0.10.0-dev.311

See the documentation for more details.