Claude usage limits can display another account's weekly quota

Ouverte
#39,770 0 commentaires 0 réactions 0 personnes assignées Voir sur GitHub

Les mainteneurs répondent en général sous 1 jour

Personne n'a encore pris cette issue.

Évaluation

Difficulté
4/5
Temps estimé
3-5 jours
Accessibilité débutants
45/100
Type d'issue
Bug
Clarté
Plutôt claire
Activité
Active
Stack technique
typescript

Piste de recherche

Start by tracing resolveUsageLimitsAfterProbe and the Claude provider usage probe and cache code named in the report. Check how authenticated account identity is obtained, how snapshots are keyed and invalidated, and how probe failures are propagated. Reproduce the mismatched-account and unavailable-response cases from the issue; done means another account's cached quota is never shown as B's fresh usage and failures remain sanitized.

Rédigé par le modèle d'indexation à partir du texte de l'issue.

Description

Environment

  • macOS (Apple silicon)
  • T3 Code Nightly 0.1.11-nightly.20261118.4864
  • Claude Code 2.1.328
  • Two or more claudeAgent provider instances using separate Claude Code configurations and macOS Keychain namespaces. Both accounts use Claude subscription OAuth.

Reproduction

  1. Set up two Claude provider instances, A and B, with distinct OAuth accounts and separate Keychain namespaces.
  2. In a disposable configuration for B, retain locally cached account metadata from A. Seed a fresh weekly usage cache whose account key also refers to A. Set the cached weekly utilization to a recognizable synthetic value, such as 37.25%.
  3. Launch a Claude Code Agent SDK probe with B's Keychain namespace and the disposable configuration. Disable tools and hooks, use an empty MCP configuration, and disable session persistence. Send initialize, then get_usage with skip_behaviors: true. No model prompt is needed.
  4. Observe the weekly result. As a control, change only the cache's account key to an unrelated value and rerun the probe.
  5. Separately, after T3 has published a weekly result for B, make the next usage probe return no limits with rate_limits_available: true and rate_limits: null, then refresh the provider. This simulates an unavailable usage response. A transient HTTP 429 was also observed during the real incident.

Expected behavior

The usage row for B should show only data associated with B's active OAuth identity. An identity mismatch or an unavailable usage response should produce a clear, sanitized error. A prior sample may be shown as stale only if it is verified to belong to the same active account. It should never appear as a fresh sample for a different account.

Actual behavior

With matching local account metadata and cache keys, get_usage returned the seeded 37.25% from A even though the Keychain credential belonged to B. The same value disappeared when only the cache key was changed. On the live T3 system, B showed A's weekly utilization and reset time. On a subsequent failed probe, T3's resolveUsageLimitsAfterProbe retained the previously published result, so the earlier attribution persisted.

The observed Claude Code cache lookup compared the cache key with local account metadata. The active OAuth credential was verified independently. The process that originally wrote the crossed metadata was not identified. No account identifiers, credential values, or local paths are included in this report.

Suggested changes

  • T3 Code: bind usage snapshots to an authenticated account identity as well as the provider instance, and invalidate unattributed snapshots. Preserve typed, sanitized failure reasons when refreshing.
  • Claude Code: bind cache reads and writes to the active OAuth account and organization. Treat disagreement between local metadata and the active identity as a cache miss.
Langage dominant
TypeScript
Étoiles
24.8k
Forks
6.4k
Merge moyen
4 h 29 min
PR mergées (30 j)
488

Préparer son environnement

Ouvrir dans Codespaces

Lance le conteneur de développement du projet dans votre navigateur, avec votre propre compte GitHub.

Par où commencer

  1. Lisez l'issue en entier, puis le guide de contribution du projet.
  2. Signalez en commentaire que vous la prenez — cela évite que deux personnes fassent le même travail.
  3. Forkez le dépôt et travaillez sur une branche.
  4. Ouvrez une pull request qui référence le numéro de l'issue.

Autres issues de pingdotgg/t3code

Toutes les issues de pingdotgg/t3code

Issues similaires

Plus d'issues TypeScript

Recevez les nouvelles issues par e-mail

Un résumé court des issues GitHub adaptées aux débutants.