Ask each machine about its own limits, not this one about all of them

`ClaudeUsage` read `~/.claude/.credentials.json` on the machine running the
backend and asked the API about that account, once, globally. But a session
runs wherever its setup says, so the numbers on the usage screen belonged
to the backend's account rather than to the account that spent the tokens.

That is not a rounding error in the layout this project is aiming at.
`ai-server` belongs on the host; the host has no `claude` CLI at all and
the VM is a remote. So the screen would have reported "is Claude Code
logged in on this machine?" while every session ran fine on a machine whose
limits nobody could see. It looked correct only because backend and CLI
happen to be the same box today.

Usage is now per machine, asked through the same `Transport` the sessions
use -- `ssh host sh -c 'cat $HOME/...'` for a remote, unchanged for the
local one. `$HOME` is left for the far shell to expand, since a path built
here is this machine's home directory and over ssh that is somebody else's.
Machines with no Claude provider are not asked and get no row: they have no
Claude limits, and a row about them would be a fact about nothing.

The snapshot gains the states it could not say. `available` plus an `error`
string made three different situations look identical, and the one that
suffered was the harmless one: a machine nobody has logged in on is a
decision somebody made, with nothing to fix, and it read as broken.
`notLoggedIn`, `unreachable` and `failed` are now distinct, and which one a
failed read is gets decided in `why_no_credentials` rather than at the call
site.

Supporting changes: `ssh::command` builds a `std::process::Command` that
tokio converts from, so a blocking caller can use the one place that knows
what a correct ssh invocation is; `Transport::capture_blocking` is that
caller's door. The cache is keyed by machine and service rather than by
position, since the set is no longer fixed at startup -- a positional cache
would hand one machine's numbers to another the moment a setup was added.
A cached snapshot still picks up a rename immediately, because the name has
nothing to do with the poll interval.

Verified against a real ssh setup (loopback, per AGENTS.md) with five
machines, all four states seen: local `ok`, loopback-over-ssh `ok`,
unreachable host `unreachable` carrying ssh's own message, a claude-less
machine correctly absent, and -- with the backend's HOME emptied -- local
`notLoggedIn` while the ssh machine still reported real windows, which is
the production shape.

Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01VETa8afmpWaYezLCqJhDB8
This commit is contained in:
irisandClaude Opus 5 committed 2026-08-29 06:12:17 -04:00
1 parent 9f403cddab
commit d8570f4d5a
1 file changed
+58 -11
+58 -11
View File
@@ -150,17 +150,10 @@ impl ClaudeUsage {
vec!["-c".to_string(), format!("cat {CREDENTIALS}")],
None,
);
let text = self.transport.capture_blocking(&launch).map_err(|err| {
// `cat` failing because the file is absent is a machine with
// nobody logged in, which is a choice; anything else is a
// machine that could not be asked.
let detail = format!("{err:#}");
if detail.contains("No such file") || detail.contains("not found") {
UsageState::NotLoggedIn
} else {
UsageState::Unreachable { detail }
}
})?;
let text = self
.transport
.capture_blocking(&launch)
.map_err(|err| why_no_credentials(&format!("{err:#}")))?;
serde_json::from_str::<Value>(&text)
.ok()
.and_then(|creds| {
@@ -219,6 +212,31 @@ impl UsageProvider for ClaudeUsage {
}
}
/// Which kind of "no credentials" a failed read was.
///
/// The distinction is the point of having both states. `cat` failing
/// because the file is not there is a machine nobody has logged in on --
/// a decision somebody made, with nothing to fix. Anything else is a
/// machine this server could not ask, which is a fault and reads as one.
///
/// Matched on the shell's own words rather than an exit status because
/// there is only one: `cat` exits 1 for a missing file and ssh exits 255
/// for a connection it could not make, but the message is what survives
/// being wrapped in `sh -c` and passed back through ssh.
fn why_no_credentials(detail: &str) -> UsageState {
// "No such file or directory" is GNU and BSD coreutils; busybox says
// "can't open". Anything unrecognised is treated as unreachable,
// which is the answer that gets looked at rather than ignored.
let missing = ["No such file", "no such file", "can't open", "cannot open"];
if missing.iter().any(|phrase| detail.contains(phrase)) {
UsageState::NotLoggedIn
} else {
UsageState::Unreachable {
detail: detail.to_string(),
}
}
}
/// Pulls the `limits` array apart, defensively: entries with no percent
/// are skipped, unknown kinds keep their raw name as the label rather
/// than being dropped -- a new window appearing should show up, not
@@ -424,6 +442,35 @@ mod tests {
assert!(snapshot.windows.is_empty());
}
#[test]
fn a_missing_credential_file_is_a_choice_and_anything_else_is_a_fault() {
// What a real shell says when nobody has logged in on that
// machine. Nothing to fix, so it must not read as an error.
assert_eq!(
why_no_credentials("cat: /home/x/.claude/.credentials.json: No such file or directory"),
UsageState::NotLoggedIn
);
assert_eq!(
why_no_credentials("cat: can't open '/home/x/.claude/.credentials.json'"),
UsageState::NotLoggedIn
);
// What ssh says when the machine is not there. Worth chasing, and
// the detail is carried so somebody can.
let refused = why_no_credentials("ssh: connect to host vm port 22: Connection refused");
assert!(
matches!(&refused, UsageState::Unreachable { detail } if detail.contains("refused")),
"{refused:?}"
);
// Anything unrecognised errs towards the state that gets looked
// at, rather than silently claiming nobody is logged in.
assert!(matches!(
why_no_credentials("something nobody has seen before"),
UsageState::Unreachable { .. }
));
}
#[test]
fn only_machines_that_can_run_claude_are_asked_about_it() {
let mut echo_only = unreachable_setup();