Gemba
Give your agent team a factory floor
The agent-runtime platform. One command family for a terminal. One set of CI actions for every push. Both run the same loop.
Every team builds the same machines again.
A team that wants to run coding agents continuously writes a bootstrap script. Then it writes a session harness. Then it needs somewhere for traces to go and somewhere for memory to live. Then it needs a way to tell real improvement from noise.
Gemba packages that work as one platform. In Lean practice, gemba is the factory floor. It is the place where value gets made, and it is the place you must stand to understand the work. This platform is that floor for your agent team. The commands are the machines on it.
Six steps. Every run leaves a record.
The loop runs stand up, then run, then see, then remember, then measure, then stop. Each step answers one question. Five steps ship as a command. The first ships as the bootstrap action and its installer.
Two of the six steps come straight from factory practice. See is genchi genbutsu. You go to the actual place, and you look at the actual thing. For an agent session, the trace is that thing. Measure asks what Shewhart and Deming asked on the factory floor. Did the process shift, or is this ordinary variation? An XmR chart separates the two.
Is the environment ready and the toolchain pinned?
gemba-bootstrapactionWhat did the agent do on this task?
gemba-harnessWhat does the trace say about the session?
gemba-traceWhat did the team learn, and where does it live?
gemba-wikiDid the metric move, or is this noise?
gemba-xmrIs the team creating work faster than a human can read it?
gemba-watchdogWhat a team rehearses locally runs on every push.
Gemba ships the loop twice. The commands run in a terminal. Four published composite actions run the same steps in GitHub Actions. A workflow pins each action by SHA.
Install the six commands, or run any one of them through
npx. A session, a trace, a memory write, and a
control chart all happen where you already work. Nothing
needs a server or a database.
gemba-bootstrap stands the platform environment
up. gemba-harness runs the session and uploads
the trace. gemba-wiki writes memory with a
freshly minted token. gemba-benchmark spreads
benchmark families across machines and merges the reports.
Gemba adds no importable API of its own. It consumes published
runtime libraries. When you need the components instead of the
commands, import @forwardimpact/libharness,
@forwardimpact/libwiki,
@forwardimpact/libxmr, and
@forwardimpact/libwatchdog directly. Read the
library catalog.
Kata runs on this platform. Daily.
Kata is an agent team that plans specs, ships features, studies its traces, and acts on findings. Its skills call five of the six commands. Its workflows pin the same four actions any other team would pin. Kata proves the platform is generic. Read the practice at kata.team.
Two defaults still name that tenant.
gemba-wiki creates a metrics directory only for a
skill whose name starts with kata-.
gemba-xmr uses kata-shift as its
default shift type. Everything else in the platform is
tenant-neutral.
Three lines to a captured trace.
Install the skill pack. Install the command family. Run one session.
The gemba-bootstrap action does the same bring-up
in CI. Its fit-install.sh installer does it on a
workstation, and that installer ships in the shared
gear release. Take the full path in
Get started, then read the
documentation.