Scenario long-tail: Claude model errors interrupt frontend review - how to retry safely or hand off client work.
Anthropic recorded a resolved incident on November 16, 2026 involving elevated errors for Claude Sonnet 5.5 and Claude Fable 5.2. The local first-party discovery record says the impact covered claude.ai, the Claude API, Claude Code and Claude Cowork during the incident window. Model and plan availability can vary by account, so the event is not proof that a particular client workflow failed.
Make the task state durable
Before retrying, record the repository revision, task identifier, model selected, input artifact, output artifact and whether a write was applied. For a code review, preserve the diff or comment state. For a generated asset, preserve the source inputs and last accepted output. A partially displayed response is not a completed deliverable.
Choose retry, switch or handoff
If the task is read-only and idempotent, retry after the provider reports recovery. If the task can write files, use a fresh worktree or reviewed patch boundary. Switch to another approved model only when client policy allows it and record the change. If neither path is safe, hand the task to an operator with the durable state record.
Use a recovery card
The practical artifact records task state, provider and model, last safe checkpoint, retry rule, alternate owner and client communication status. It makes the provider incident one input to the decision, not an excuse to overwrite uncertain work.
Applicability and limits
This applies to frontend review, build or content workflows that explicitly used the affected Claude surfaces and models during the incident window. It does not establish data loss, customer impact, an API-wide failure outside the status record or a need to change provider settings. No API call, repository or client workflow was exercised for this article.