You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
{{ message }}
Repository navigation
Commit a97b247
Browse filesBrowse the repository at this point in the historyBrowse files
feat(muse): show Muse Code token usage and subscription limits
The Usage section only showed Codex. Muse Code had no token history
reader and never reported subscription limits, so neither the Tokens
view nor Limits mentioned it.
Mark Muse's full two-window report explicitly. Complete reports replace
stale windows and clear a failed-probe notice; partial live updates
preserve upstream's notice behavior.
Add a "muse" usage kind with a transcript reader over the native session
logs: model completions and compaction summary metrics contribute
counted-once input, output, and reasoning at their recorded time, with
record-id deduplication across forked session files. Summary metrics
omit the model, so retain their tokens as unknown and unpriced. Rescan
existing Muse files from older usage caches while preserving deleted
history. Read Muse's MSP usage/read during the existing read-only status
probe and map usage/changed notifications from turn hosts to session and
weekly limit windows, so Limits, /usage-limits, and the mobile widgets
show Muse beside the other providers. A fresh probe host has observed
nothing, so Limits first shows an honest not-reported notice and keeps
the last observed windows after turns report.
Scope the credential fingerprint to hostname, canonical home path, and
filesystem identity so one login pools once without merging identical
paths on other machines. Preserve fingerprints across window-only
updates, advance stale checkedAt at most once a minute unless reset
credits ride the snapshot, and reject record timestamps aggregation
cannot bucket.
Implemented by GPT-6-Astra at Max, reviewed by GPT-6.1-Sol at Extra High
across three rounds, led and verified by muse-wonder-full-1.0 in Muse
Code via T3 Code.
Summary accounting validated with 102 focused tests, scoped typechecks,
lint, and a read-only scan of 39 native summary records. Implemented by
GPT-6-Astra through Codex; reviewed by GPT-6.1-Sol at Extra High.
0 commit comments