stet

stet is a read-only companion TUI for inspecting an agent's changes.

The agent runs in one terminal pane, but you still open an editor just to answer basic questions:

• What files are in this repo?
• What changed?
• What did the agent touch most recently?
• Are there errors or warnings in what changed?

stet is meant to sit in the next pane and answer those questions without becoming part of the agent loop. It does not review code, approve changes, talk to the agent, or manage a workflow. It shows you the repo, the diff, and the problems. You decide what to say next.

Stet is the proofreader's mark for "let it stand": strike a word out, add dots beneath, and it stays.

What it does

• Shows the full repo tree, including tracked files and untracked files that are not ignored by git.
• Marks changed files in place, with staged, unstaged, mixed, and untracked states.
• Opens unchanged files read-only, with syntax highlighting for any language Shiki supports.
• Opens changed files as diffs, with a toggle for the full file.
• Shows an image or binary file as a metadata card (type, dimensions, size) instead of raw bytes, since a terminal can't render the pixels, and opens it in the OS default app with O.
• Finds text within the open file, and searches file contents across the repo, scoped to the changes or the whole tree.
• Switches scope from a picker (all changes, staged, unstaged, since launch, or the last commit) and between git worktrees in place.
• Shows which worktree an agent is working in: the picker ranks worktrees by how recently each was touched (an edit, a commit, a checkout), busiest first, each with its age beside it.
• Watches the filesystem and refreshes as the agent changes files, keeping the current file and selection stable.
• Marks recent activity and lets you jump to the latest touched file.
• Shows per-line provenance with a: a gutter rail that places each line on a five-tier timeline, uncommitted, committed this session, committed on this branch, changed since the file's start, or original to it, git blame reframed around how much of what you are reviewing is yours and how recent it is.
• Shows diagnostics in the tree, the viewer, and a problems panel.
• Navigates code through read-only language-server pulls: go to definition, find references, find implementations, call hierarchy, hover, and a symbol outline.
• Copies a reference and snippet to paste back to the agent: path in the tree, path:line:col in the viewer, and a whole diagnostic (or every one) from the problems panel.

The git-backed file tree renders first. Diagnostics come in later as decorations, so the basic view stays useful even when checks are still running.

Install

# standalone binary (macOS / Linux, no runtime needed)
curl -fsSL https://raw.githubusercontent.com/jimmy-guzman/stet/main/install.sh | bash

# npm (works with npm, bun, pnpm, yarn; pulls a prebuilt binary)
npm i -g @jimmy.codes/stet

# homebrew
brew install jimmy-guzman/tap/stet

Run stet inside any git repository; stet upgrade updates it in place.

Documentation

Full documentation lives at stet.jimmy.codes:

• Getting started: install, usage, flags, and requirements
• Guides: reading files and diffs, search and navigation, code intelligence, scopes and worktrees, and themes
• Keybindings: every shortcut (or press ? in the app)
• Configuration: themes, file associations, icons, language servers, and editor setup at ~/.config/stet/config.jsonc (or config.json)

Requirements

• git
• a clipboard tool for copy (y): pbcopy on macOS (built in), or wl-copy, xclip, or xsel on Linux
• a Nerd Font for the tree's file-type icons (optional; use --no-icons without one)

Development

bun install
bun run stet             # run from source (the TUI lives in packages/tui)
bun run check            # format, lint, tests, typecheck + knip
bun run build:dist       # build standalone binaries for all targets

bun install also wires up git hooks (via lefthook): pre-commit formats and lints staged files, pre-push typechecks the whole repo and runs knip.

Non-goals

stet is deliberately not an agent integration. It has no approvals, no accept/reject protocol, no generated review explanations, no PR workflow, and no database. The agent never hears from stet, only from you.