Commands/Cookbook/Troubleshooting: three claims were wrong, and half the object was undocumented
A doc review against the actual output caught them:
auth.kind "none" was glossed as "a stdio child". It is also every http server
with no auth configured, so anyone reading kind===none as "stdio" mislabels
their remotes. transport answers that question.
"A revoked refresh token still reads as valid" was only half true, and
Troubleshooting turned the other half into bad advice: once the access token
lapses the same file reads refreshable, which is where a revoked session spends
most of its life. Both places now name both states.
"Daemon down means every server idle" — a disabled server stays disabled.
The fix-list one-liner ignored .enabled in all three copies, so it told you to
re-auth a server you had deliberately turned off. Now filtered, and the field
table says why.
The field table also documented roughly half the object it is cited as the
reference for: name, transport, note and every top-level field were missing,
along with the auth.state glosses that mduct help already had. Plus the new
error field, the exit-code contract, that CLI tools are not in this output, the
idle TTL number, and that --json here is not the payload-stripping --json of
call (that one belongs on Arguments & output, which now says so).
Anonymised the example's uid and expiry — they were the author's real ones.
Commands/Cookbook/Troubleshooting: the state as data, and what it cannot tell you
status --json and servers --json exist now, so the pages that told people to
read padded columns get the field reference, the jq one-liner that names the
servers needing a login, and the two honest limits: idle is the resting state,
and a revoked refresh token still reads as valid.
Commands: --name=value, unknown options, --version
Commands: full reference — every command, its flags, exit codes