DEV Community

Cover image for Eisen Flow: Yet Another To-Do App But Better
Mohammed E. MEZERREG
Mohammed E. MEZERREG

Posted on

Eisen Flow: Yet Another To-Do App But Better

GitHub Copilot CLI Challenge Submission

This is a submission for the GitHub Copilot CLI Challenge

What I Built

This is an idea I got since I was working with my previous employer. We were using Jira for managing projects. In parallel of Jira I used Toggl, Trello, Clockify and even the Pomofocus timer.

I used these tools both through web and mobile. None of them satisfied me. I am not saying they're bad, but they didn't have what I was looking for.

I took this challenge to test a concept of my idea and see if people like it and why not win the challenge :) .

The idea revolve around implementing the Eisenhower matrix in a task or to-do list app. I didn't have the time to implement everything. But the MVP of the app is here.

Demo

You can check the app here.

My Experience with GitHub Copilot CLI

I used Copilot CLI for everything. I only used NeoVim to change some lines every now and then. You can say I developed this project through prompting Copilot CLI (Haiku, Sonnet, and Opus) 98% of the time.
I started with a free plan with Haiku. After three sessions I think I hit the limit and I was asked to wait until March. Since I have to finish the MVP and share it here. I was obliged to go buy a month of the pro plan.

Since then, I selected Opus and went through the rest of the plan finished it in one set. Then I added more features to implement to the plan and gave Copilot all permissions in the project's folder.

I changed the direction for this app two times and the agent never failed me. It even remembered removed changes even though I wasn't in a git repository! I don't know how is that possible to be honest.

i discover that I Copilot is more mature than ever. It has everything other tools have. It has skills, plan, MCPs, and so on. The funny thing is me starting a new session every time I start my machine. To my surprise they have a continue/resume command as well. So for the last three days I used

/session rename eisenflow
Enter fullscreen mode Exit fullscreen mode

to rename the last session for this app. Then every time I want to work on this project I just use:

copilot --continue
Enter fullscreen mode Exit fullscreen mode

This when I am sure that I didn't use Copilot for something else. It resumes the last opened session.
To select a session or provide the ID of a session I use:

copilot --resume # A list will show and I select my session.
copilot --resume session-id # I never used it
Enter fullscreen mode Exit fullscreen mode

This really helped and gave me a GUI vibe. It's similar to selecting an existing thread from Antigravity, Cursor, etc. Or select a chat from ChatGPT, Claude, etc.

Some Usage Stats

Sessions by log size

  1. Feb 12 — 132 KB / 1,550 lines (Pomodoro + Task UX)
  2. Feb 3 — 104 KB / 1,228 lines (Core app build)
  3. Feb 10 — 84 KB / 969 lines (Feature sprint)
  4. Feb 2 — 80 KB / 898 lines (Scaffolding)

🤖 AI Model Requests

Date Model Requests
Feb 1 0 (setup only)
Feb 2 244
Feb 3 346
Feb 8 (AM) 33
Feb 8 (PM) 29
Feb 10 288
Feb 12 464 ← heaviest
Feb 13 10
Total 1,414 model requests

📐 Context Window

Metric Value
Context windows used 128K tokens (primary), 90K tokens (sub-agents)
Peak utilization 80% (~102K/128K tokens) on Feb 10 & 12
Compactions triggered 0 (never hit threshold)

🕵️ Sub-Agent Usage

Type Count
Explore agents 3 (all on Feb 12)
Task agents 0
General-purpose agents 0

⏱️ Session Durations

Date Duration
Feb 1 22 min
Feb 2 4h 15m
Feb 3–5 ~50h (long-running session)
Feb 8 (AM) 12h 9m
Feb 8 (PM) 12 min
Feb 10 11h 9m
Feb 12 19h 15m
Feb 13 - 15 38h 36m 38.398s
Total ~136 hours of session time

💬 Prompt & Turn Counts

Metric Count
User prompts 91
Assistant turns 1,406 (~15.5 turns per prompt)
Tool executions 1,563
Context compactions 8

🧠 Model Usage

Model Role Calls Logged
Claude Opus 4.6 Primary model (premium ⭐) ~1,414 requests (main agent)
Claude Haiku 4.5 Sub-agent (explore/task) 55 calls
Claude Sonnet 4.5 Sub-agent (general-purpose) 2 calls

⚠️ Quota Issues

Date Event
Feb 8 10 quota exceeded errors ("You have no quota")

📊 Estimated Premium Requests

Source Est. Count
Main agent (Opus) ~1,414
Sub-agents (Haiku/Sonnet) ~57 (non-premium)
Total premium requests ~1,414

Well toward the end GitHub Copilot CLI advised me to use general purpose models as he refers to Haiku and Sonnet to save on premium models requests like Opus. I took his advise and worked with Sonnet for quite a bit after that.

Tech Stack

The stack used in this project is:

  • React/Vite with TypeScript of course.
  • Convex for DB, Auth and so on.
  • Cloudflare Workers for hosting.

Top comments (0)