anthropics / anthropics/claude-code
Auto-mode safety classifier intermittently unavailable for Write/Bash/Edit, blocking work for extended periods
- Vorherrschende Sprache
- Python
- Sterne
- 145k
- Forks
- 23.1k
- PR-Merge-Kennzahlen
- PR-Kennzahlen ausstehend
Beschreibung
During a long working session, the auto-mode permission classifier repeatedly returned "claude-sonnet-5 is temporarily unavailable, so auto mode cannot determine the safety of Write/Bash/Edit right now" for basic file-write and shell actions. This happened in clusters: one stretch of ~5 consecutive failures across Write, Bash, and Edit over roughly 15 minutes, requiring multiple manual retries (including a 10-minute wait) before the classifier recovered on its own. No content in the blocked commands was unusual (writing a plain Python script, editing a string in a file) -- the failures appear to be classifier-service-side unavailability, not a legitimate safety determination. This significantly disrupted a debugging workflow. Requesting investigation into the classifier service's reliability/uptime, and ideally a faster fallback or retry behavior instead of a hard block when the classifier itself is down.
Beitragsleitfaden
Für dieses Repository ist kein Beitragsleitfaden indexiert
Rechercherichtung
Start from the auto-mode permission classifier path for Write, Bash, and Edit actions, then try to reproduce the reported temporary-unavailable failures during basic file writes and shell commands. The issue names no files or tests, so the first task is locating that classifier and its retry/fallback handling. Done means classifier-service unavailability no longer hard-blocks work for extended periods, or there is a clear reliability fix with coverage.
Vom Indexierungsmodell aus dem Issue-Text verfasst.
Bewertung
- Tech-Stack
- bash, python
- Bereich
- cli, observability, security
- Issue-Typ
- Bug
- Schwierigkeit
- 4/5
- Geschätzter Aufwand
- 3-5 Tage
- Aktivitätsstatus
- Aktiv
- Klarheit
- Größtenteils klar
- Anfängerfreundlichkeit
- 35/100