The latest articles on DEV Community by Praful Reddy (@prafulreddy). https://dev.to/prafulreddy https://media2.dev.to/dynamic/image/width=90,height=90,fit=cover,gravity=auto,format=auto/https:%2F%2Fdev-to-uploads.s3.amazonaws.com%2Fuploads%2Fuser%2Fprofile_image%2F3871637%2Fdfb49680-d4a7-46ca-b39b-0859d4c61fbd.jpg https://dev.to/prafulreddy en Praful Reddy Fri, 10 Apr 2026 11:48:59 +0000 https://dev.to/prafulreddy/a-zero-token-progress-bar-for-claude-code-51bp https://dev.to/prafulreddy/a-zero-token-progress-bar-for-claude-code-51bp <p>Every Claude Code extension I've seen shows the same thing: token counts, API costs, model info. Useful, but it doesn't answer the question I actually care about mid-session — how much of the work is done.<br> So I built task-progress-bar. It reads Claude Code's native task list from disk and renders a live ASCII progress bar with time estimates. It runs as a PostToolUse hook, which means it consumes zero tokens — it's a subprocess that Claude never sees.</p> <p>Tasks [████████░░] 8/10 (~3m left) | ✓8 ⟳1 ○1</p> <h2> How it works </h2> <p>Claude Code persists tasks as JSON files in ~/.claude/tasks/. The plugin watches for TodoWrite, TodoRead, TaskCreate, and TaskUpdate tool calls via the PostToolUse hook, then:</p> <ol> <li>Parses every JSON file in the tasks directory</li> <li>Counts completed, in-progress, and pending tasks</li> <li>Computes a time estimate using an exponential moving average (EMA)</li> <li>Outputs a single status line to Claude Code's statusLine renderer</li> </ol> <p>The time estimation is straightforward. Each time a task moves to completed, the timestamp is logged. The interval between consecutive completions feeds into an EMA with α=0.3:</p> <p>EMA_new = 0.3 × latest_interval + 0.7 × EMA_old<br> estimated_remaining = EMA × tasks_left</p> <p>Intervals over 1 hour are clamped to avoid skew from session breaks. The first 3 completions show "calculating..." until there's enough data.<br> The entire plugin is a single Python file — stdlib only, no pip dependencies. It uses json, pathlib, time, and sys. Nothing else.</p> <h2> Install </h2> <p>curl -fsSL <a href="https://raw.githubusercontent.com/PRAFULREDDYM/task-progress-bar/main/install.sh" rel="noopener noreferrer">https://raw.githubusercontent.com/PRAFULREDDYM/task-progress-bar/main/install.sh</a> | bash</p> <p>The installer checks for Python 3.8+, copies the script to ~/.claude/, and patches settings.json with the hook configuration. There's a matching uninstall.sh for clean removal.</p> <h2> What it looks like </h2> <p>The progress bar color-codes by completion percentage — red below 33%, yellow from 33–66%, green above 66%. When all tasks finish, it shows ✅ All done! for 30 seconds and then hides.<br> If you run it standalone (python3 task_progress_bar.py), you get a full multi-line colored view with a task breakdown and per-task average.</p> <h2> Links </h2> <ul> <li>GitHub: <a href="https://dev.tourl">github.com/PRAFULREDDYM/task-progress-bar</a> </li> <li>Requirements: Python 3.8+, Claude Code v2.1+</li> <li>License: MIT</li> </ul> claudecode productivity terminal python