Stepwise — live progress for terminal coding agents

Give Claude Code and other terminal coding agents a per-step shell command progress view

Problem

When a terminal coding agent runs shell commands, users only see a long spinner—no idea which step it's on, which command it's stuck at, or whether it already failed. The wait is opaque and failures require digging through logs.

Solution

A lightweight sidecar that reads the agent's shell command stream and shows each command's timing, status and output summary in real time, with failure highlighting and replay, making terminal coding observable.

Users

Developers who use Claude Code, Cursor or other terminal/IDE coding agents daily

MVP

Capture and parse the agent's shell command stream; show each command and duration live; highlight failed steps; replay view for the last N sessions

Revenue

Open-source core plus Pro subscription (persisted history, parallel multi-session view, shared team dashboard)

Why now

Terminal coding agents are spreading fast but their execution is still a black box; a sidecar observer doesn't touch the models and is very feasible for an indie