You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
The sidebar fits far more sessions. Each one is a single row, as in Claude Code: its name, then how long ago you used it. Point at a row for its ⋯ menu; the exact time is in its tooltip.
Fewer approvals
Commands that only look run at once in Ask before changes: ls, cat, grep, rg, find, git status, git diff, git log and the like, joined with |, && or ;, as long as every path they name is inside the folder. Anything that writes, runs something else or leaves the folder still asks. The README has the full list.
In PowerShell too, on Windows without Git for Windows: Get-ChildItem, Get-Content, Select-String, Sort-Object, Where-Object, Measure-Object and the same git commands. ForEach-Object, { }, $variables, env: and the registry still ask.
Secrets still ask, even in Auto-approve, however the command spells them:cat .en""v, gc .en''v and git show HEAD:.env ask like cat .env. Reading through a wildcard that could match a secret (cat .env*, Get-Content *.pem) is never counted as only looking.
And
Give us feedback, a link under the composer, sends a few words to the CodingFleet team with the app's version and system, never a chat or a file.
The model picker says Unlimited for a model as soon as one of its efforts is included in your plan, instead of showing a credit range; the effort menu still prices each effort.
The agent can wait up to 4 minutes for a background command in a single read, instead of 30 seconds, so a long build no longer means checking over and over.