PhaseDrift

Set up with your agent

Connect PhaseDrift to the project you work in.

Choose report analysis

Managed analysis uses your PhaseDrift subscription. BYO uses your TypeSafe key. Both are optional and require separate consent after you inspect the report payload.

You can connect local reporting first. A license or provider key alone does not authorize uploads. Choose your analysis setup.

Copy this into your coding agent

Open your project, copy the prompt, and paste it into the agent. It handles installation, project instructions and reporting.

Read the complete setup prompt
Open prompt as plain text

Need the CLI? Install the signed release first. The skill download adds instructions; it does not install PhaseDrift.

How you know it worked

Open the dashboard URL your agent returns. Check the project name, then find its setup report in Agent work → Reports.

The first code scan may still be running. See its progress in Code health; you can keep working meanwhile.

Start your next task

Reload the project instructions or start a new agent conversation. Finish one ordinary task, then look for its new report in Reports. The installed instructions tell the agent to report without a reminder from you.

No new report? Ask your agent:

Check PhaseDrift reporting in this project. Run phasedrift doctor, confirm you loaded the installed reporting skill, and verify the last task's report was saved. Fix the connection if needed. Preserve the existing data.

Your next useful request

Use PhaseDrift to find one repeated obstacle in this project. Show the original reports or check results, propose one small fix, and explain how to check it. Wait for my choice before implementing it.

Try one useful change →

Setup reference

Jev setup and request limits. Your agent checks the chosen transport, consent and limits. Reports and proposals work while analysis is off.

Install manually · Project skills · Choose your coding agent

Plain-text setup prompt · Machine-readable prompt

Default setup connects the repository to your account and syncs code-health and agent-work summaries, including future measurements. Paid analysis stays off. Git worktrees share one project and computer-level store. Ask your agent to verify both an acknowledged upload and the real project in your signed-in account. Choose local-only setup explicitly when you want all history on this computer.

Multiple worktrees

A Git worktree is another checkout of the same repository. PhaseDrift keeps its history outside those checkout folders and uses the shared Git repository to find it. All worktrees therefore appear as one named project in your account. Each checkout keeps its own branch, files and command working directory. Run phasedrift open from any registered worktree to open that project.

The cloud name comes from the repository name, rather than a folder such as worktree-7. Existing connected projects publish their name on their next sync after an upgrade. Syncing a name does not resend measurements or change project identity.

Give merged code its own baseline

After account sync works, ask the setup agent to follow GitHub baselines. The App connects the exact repository; the pinned workflow measures its default branch. The agent verifies runner cost and permissions before enabling it. Local effort and feedback remain separate.

On this page