Quickstart
From zero to your first AI review in about 60 seconds.
From inside the git repo you want to review:
1. One-time setup
Writes the nine /staff-* skills, creates the .staffreview/ store, and
gitignores the per-machine bits.
staff install2. Open a diff in the browser
WT is the working tree; STAGED is the index.
staff main..WT3. Review and resolve in your agent
In your AI agent (e.g. Claude Code, from the same repo):
/staff-review main..WT # leaves inline comments — they show up live in the UI
/staff-resolve # fixes / documents / skips each open commentWatch the comments stream into the browser as the agent posts them. Triage in
the UI, reply, resolve threads yourself, or let /staff-resolve work through
them. That’s the whole loop.
Review against your working tree
Use a …..WT diff (like main..WT) so that fixes from /staff-resolve flow
into the next review. This is required for /staff-loop to
converge — point it at the working tree, not a fixed commit range.
Hands-off: let the loop run
Once you trust the setup, chain review and resolution automatically:
/staff-loop main..WT/staff-loop runs /staff-review → /staff-resolve in subagents, round after
round, until a fresh review surfaces nothing new (or it hits the
loopMaxRounds cap — default 5, adjustable in the gear menu). It edits your
working tree but never commits, so you read the diff before committing.