I'm Pavel, the founder of DevGhost. This post walks through what I built it for: an engineering lead who lives in VS Code and wants to understand what their team's Git history says about the work behind it.
Commit counts and lines of code don't answer that. A one-line fix can take a day; a generated migration can add 2,000 lines in a minute. So DevGhost estimates something else.
The baseline
For every change, DevGhost estimates how long it would take a reference developer: mid-level, familiar with the codebase, working without AI. That gives every person and every period the same yardstick.
Two numbers come out of it:
- Contribution, h: estimated reference-developer hours behind the changes. Not time actually worked.
- Ghost%: contribution per working day relative to the reference developer. 100% means the reference developer's daily output (the methodology uses 3 productive, focused hours a day). Someone working with a strong coding agent can be well above 100%.
Neither number is a quality score or a verdict on a person. They tell you where to look.
Start with the team
Install the extension, open the Command Palette, run DevGhost, and choose Try the sample. It opens Ghost Café, a fictional team bundled with the extension. It's free and works offline.
The view I start with is Effort and Ghost% by person with Recommendation zones turned on. The horizontal axis is contribution in hours, the vertical axis is Ghost%. The dashed vertical line is the team's median effort; the horizontal one is the reference developer at 100%. That splits the team into four zones, each phrased as a conversation starter rather than a grade:
- Recognize: high effort, above the line. Recognize the contribution and watch for overload.
- Unblock: high effort, below the line. Understand the work and clear blockers.
- Grow: efficient but less total effort here. Room for more scope, or time spent elsewhere.
- Check: low activity here. Confirm where their work actually is before reading anything in.
The chart itself says what it can't see: review, mentoring, design, and non-coding work.
Then one person
In the sample, Vince Breakpoint has the largest bubble on the team (8.95 estimated hours over the period) but sits in Unblock: his Ghost% is 49.7%. Over 6 working days that's about 1.5 reference hours a day, half the baseline.
That doesn't tell you Vince is slow. It tells you there's a question worth asking. Open his card: daily contribution against the reference level, with the Ghost% line over the selected period.
Then the commits
From there you go down to the individual commits behind the estimate. This is where the real conversation happens: was it a hard change in an unfamiliar module, a long review cycle, or work that isn't in this repository at all? The estimate is a starting point; the commits are the evidence you discuss with the person.
What else is in there
- Branch and period selection, team trends, and Replay of how the chart evolved.
- Merging several Git identities that belong to one person, marking bots, excluding people from the report.
- Export to PDF, Excel, or JSON.
Local by design
All analysis and analytics run on your machine. Your code and results stay there. Installing, updating, and buying extra credits need a connection; analysing history doesn't.
Try it on your own repository
DevGhost for IDE runs in VS Code 1.95+ and Cursor on Windows x64 and macOS 14+ (Apple Silicon and Intel). You need Git and a local clone.
- Install from the VS Code Marketplace or Open VSX for Cursor.
- Open the sample to learn the views.
- Choose Add repository, pick a period you know well, and compare the result with what you know about your team.
500 new commits are free every month. Beyond that there are one-time credit packs, no subscription.
I'd like to hear where the estimate matched your sense of the team and where it didn't. Those mismatches are what I'm working on.
Top comments (0)