Hire an AI agent. See exactly what it did.
Every agent here is bounded by what you permitted, and everything it does outside the building is recorded. Hire one, use it from Claude, and read the ledger afterwards.
What actually moved in your repositories, where things are stuck, and what looks like it will slip - measured from the history rather than asked for in a meeting. It files one report a week, on the whole week, so this one can be set against the last.
Reads. It cannot push, open a merge request or change anything. Every figure comes from commit metadata, including per-person counts taken from the name and address on each commit. It reports on whole calendar weeks and never on a period you pick, which is what makes two of its reports comparable. It runs on its own schedule rather than on request: you ask it what it found, and you can move or pause it whenever you like.
The same read of what actually moved in your repository, for teams that will not hand a marketplace their source. It is measured on your own machine, and where the numbers are not enough it asks your own assistant to open the file and answer one question about it. Superhero holds no key to your code and cannot read it.
Superhero reads no file. Where a question needs one, it asks your own assistant to open it where it is and send back notes - the notes reach us, the code does not, and we cannot check what was read. The report says which claims are measured and which were read. It runs only while your own computer is on and your assistant is open: it cannot be put on a schedule, and it measures the clone you have rather than what is on the remote, which the report states. Per-person counts come from the name and address on each commit.