Which version am I on
~/.base-gbl/manifest.toml. That is tracked separately and can drift, most easily by swapping the binary by hand without touching the manifest. When the two disagree, base --version is the truth and the manifest is the thing to fix.
Is there a newer one
base binary is checked against GitHub releases; other ChrisAI components are checked against npm.
Updating
--force reinstalls even when you are already current, which is the way out of a state where the manifest and the binary disagree.
Quieting the banner
manifest.toml, which holds the installed version, the install path and timestamp per component, plus an update-check block with when it last checked, how long between checks (7 days by default), any pending update, and the snooze expiry.
The gotcha. That single file being the source of truth is exactly why a hand-swapped binary makes the banner nag about a version you already have. base update --force rewrites it.
What an update does not do
This matters more than it sounds, because the skill is the answer surface. Its answer bank stamps itself with the version it was checked against, and it is written to treat every answer as a lead to re-verify when your installed version does not match that stamp. So a stale skill degrades honestly rather than lying to you. It still degrades. After any update, runbase install again. It takes seconds and it puts the coach back in step with the binary.
After an update
commands.toml that stops parsing takes every star command in that tier with it silently, and base doctor is what tells you.
Next
Coach mode
The bundled skill, and why keeping it current matters.
Known issues
Behaviour that will surprise you, checked against the version that ships today.