Connect your first project
Choose your install path. Connect a repo. Check the result.
Prefer your agent to handle it? Use the setup prompt →
Get PhaseDrift
You need macOS or Linux, Git, Bun 1.3+, Node.js, and Perl. Node runs additional analyzers; Perl runs the bundled code counter.
Check that your terminal can find it:
phasedrift --versionIf the command works, connect your project below. If it is missing, choose your available installation path above.
Installed, but the command is missing?
Add the install location to your terminal's path:
export PATH="$HOME/.local/bin:$PATH"Add that line to your shell configuration for new terminals. Then retry phasedrift --version.
Connect a Git repository
Run this inside the project you want to understand. Replace the quoted example path with yours:
cd "/path/to/your/project"
phasedrift init
phasedrift openinit starts the tracker and installs the project operator and reporting skills, managed instructions and supported project hooks. Repeat it to reuse the connection and data. Reload your agent's project instructions afterward. The downloadable kit is optional for manual installation and reference.
Confirm you're connected
phasedrift status
phasedrift doctorCheck the returned project name and dashboard URL. Use that URL, not a fixed port from an example. doctor gives the next remedy when something needs attention.
You are connected when status names the correct repository and returns its dashboard URL. Open that URL.
The initial code scan runs in the background. Code health shows its progress and any tool that needs attention.
What the first scan collects
Setup queues a full working-copy scan, the latest 10 commits, and up to 20 older history samples. Follow progress in Code health. Some analyzers need project dependencies or another executable.
Sampled history shows the shape of change over time. It does not run every full check on every commit. Blocked or unavailable tools remain partial.
Do one normal task
Load the installed project instructions in your agent. When the task finishes, open Agent work → Reports and find its actual task, checks and timestamp. This is your reporting success check; a populated chart alone does not prove it.
Check host coverage for native effort measurements and reporting identities for other hosts. Missing measurements stay unavailable.