The agent hands off
At the plan gate or after a diff, the agent opens a cueloop review and waits for your verdict instead of charging ahead.
cueloop opens your agent's plan or diff in the terminal, where you annotate the lines and send back a verdict it acts on.
npm i -g cueloop» cueloop plan Add rate limiting to the API plan · pending 1 Add a token-bucket limiter in middleware/ratelimit.ts 2 Return 429 with Retry-After on overflow 3 Store counters in Redis (a new dependency) ▌ alex ▌ per key or per IP? keys rotate, so IP is ▌ the safer limit against abuse. [ comment ] approve request changes q · quit
Your agent proposes a change and waits for your call. You read it in the terminal and reply with a verdict it can act on. Nothing leaves the command line.
At the plan gate or after a diff, the agent opens a cueloop review and waits for your verdict instead of charging ahead.
Read it in place. Select a line, quote it, and attach a comment or a suggested change, anchored to the exact text.
Comment, approve, or request changes. Your notes become a structured feedback document the agent reads and acts on.
One verb per artifact. cueloop plan opens the latest
plan an agent proposed. cueloop diff reviews the working
tree. cueloop review pulls a pull request into the same
surface and posts the verdict back to the PR.
» cueloop diff M ratelimit.ts +42 -3 M server.ts +8 -1 + limiter = tokenBucket(100) // per key / min + return 429 on overflow + store counts in redis - // TODO: rate limiting [ comment ] approve request changes
Press Share and cueloop copies one line. Your
teammate pastes ssh <id>@cueloop.dev and the plan
opens in their terminal, your annotations already on it.
Their notes come back to you attributed to them, over SSH the whole
way.
» ssh [email protected] the plan opens in your terminal, annotations already in place. Add rate limiting to the API shared ▌ alex per key or per IP? ▌ keys rotate. ▌ you good catch, let's key on IP. [ comment ] submit annotations
Comments attach to the exact lines they are about, and follow along as the plan or diff changes underneath them.
Your agent pauses for your review, then picks up your feedback and keeps going. Built into Claude Code and Codex.
Nothing to host, no account to create. cueloop runs wherever your shell does.
Share a review and everyone's comments come back in one place, attributed and intact.