CobaltCode (@CobaltCode_ai) header
CobaltCode (@CobaltCode_ai) profile picture

Give your coding agent a dedicated computer for each task. Run your app, preview your changes, create a PR - across as many tasks as tou want

Joined July 2026
Review a result from your phone. Switch between the conversation and Preview and your half-written message is still there when you come back.
Group your tasks by repository. Expand the project you are working on, collapse the rest, and reach last week's task without paging past a busy review Loop.
1
Three improvements to the Cobalt workspace: repository groups in the task sidebar, a mobile layout with more room for chat and Preview, and unread messages in your browser tab.
5
Ask for the check and the evidence together: "Write and run a Playwright check in Agent Browser that saves the display name, reloads, and asserts it persisted. Record the run with captions and show me the script and results."
1
7
One script runs the whole journey: fill the form, save, reload, assert. Your agent spends its next turn reading the result instead of asking for every click.
2
Cobalt agents can now run Playwright scripts in Agent Browser. Ask for a repeatable check of your application and get back the assertions and a captioned recording.
1
Give Bug triage a report it can test: "Reproduce the page-count error: 21 items at ten per page returns two pages instead of three. Add a regression test. If the fix is localized, implement it and open a PR with the before-and-after results."
2
Bug triage reads the issue and its discussion, finds the likely cause, and posts an implementation plan before it changes any code.
New guide: automate bugfixes with Cobalt Loops. Follow one pagination bug from the report through investigation, a tested correction and an independent review.
2
Teach a Common Loop about your repository: "Update this PR review Loop to check CSV imports for duplicate customer rows and interrupted uploads. Run the import tests and report the exact failing case."
1
PR review keeps one reusable task per pull request. When the author pushes a new revision, the next review already knows what it found last time.
1