A fast and minimal coding agent, for the terminal

$ curl -fsSL https://raw.githubusercontent.com/scmmishra/clue/main/install.sh | bash

v0.3.0 (notes), macOS and Linux, status: early

Three waits, measured

Start to a ready prompt: 7.3 ms
Resume to your last session: 8.7 ms
First token, clue's time: 5.2 ms

Medians on one Apple silicon Mac. The first-token figure uses a local stand-in for the API, so it shows only clue's own time. benchmarks

~/code/app

$ clue
clue 0.3.0 · /help for commands

┃ the conversation spec is flaky, fix it

  I'll run the spec a few times to see how it fails.

● 3 tool calls · 1 command · 1 read · 1 search
├ Ran bundle exec rspec spec/models/conversation_spec.rb
├ Read spec/models/conversation_spec.rb  180 lines
└ Searched "last_activity_at"  12 matches

  The spec compares `last_activity_at` with `Time.now`. The two
  timestamps can land in different seconds. I'll freeze time in that
  example.

● 2 tool calls · 1 edit · 1 command
├ Edited spec/models/conversation_spec.rb  +3 −1
└ Ran bundle exec rspec spec/models/conversation_spec.rb

  Fixed. The example now freezes time, and 5 runs in a row pass.

  41s · $0.21 · 48k in · 1.2k out

┃

auto · opus 5.5 · medium                               ~/code/app (main)

Ready before it's loaded

clue shows the prompt first. The SDK, MCP servers and skills load while you type.

Resume without the wait

clue -c prints only the last screen of the conversation. While you type, clue sends the history to Claude ahead of your message.

A prompt Claude has already seen

The system prompt and tool list are the same bytes in every session, so Claude caches them across sessions.

A quiet transcript

Each run of tool calls shows as a short tree. Output, diffs and errors wait behind ctrl+o.

Three permission modes

auto runs routine work and has risky actions reviewed first. ask asks before each action. full-access runs every action without checks.

Your instructions

clue reads AGENTS.md and CLAUDE.md files. Projects set up for Claude Code work as they are.