Playbooks · Coding tasks
Keep coding when Claude sign-in is down
Written November 15, 2026
Two Claude incidents on November 13 split sign-in from the API. A spare Console key prepared early, and the routine audit to run afterwards.
Two incidents, two different shapes
November 13 brought two separate Claude incidents, and they were not the same failure twice. The first incident on Anthropic's status history opened at 09:58 UTC with a narrow report: an issue with users authenticating to Claude.ai and Claude Code. It put claude.ai into major outage, Claude Code and the Claude Console (platform.claude.com) into partial outage, and resolved at 11:58 UTC. The Claude API is not among that incident's affected components. The second incident opened at 13:09 UTC the same day and was wider from the start: elevated errors on claude.ai, the Claude API, and Claude Code, with all three in partial outage at points before it resolved at 15:38 UTC. The practical lesson sits in the difference. A morning in which sign-in is down but the API is untouched rewards a completely different preparation than an afternoon in which everything is degraded at once.
Why the spare key has to exist before the outage
The fallback for the first shape is a Console API key you already hold. Anthropic's Claude Code authentication documentation describes the Console route: an organization uses a Claude Console account, invites users, and assigns roles — the Claude Code role can create only Claude Code API keys, while the Developer role can create any kind of API key. A key created on a quiet day sits ready. A key you try to create during the incident may not be creatable at all: in the November 13 morning incident, the Console itself was one of the components in partial outage, and it only returned to operational in the 11:05 UTC monitoring update, after Claude Code did. Preparing the key early is not belt-and-braces caution; it is the difference between a fallback that exists and one that depends on the very surface that is down.
How Claude Code uses the key
The mechanics are simple, and the authentication documentation spells them out. If you have set the ANTHROPIC_API_KEY environment variable, Claude Code can skip the login prompt: on first launch it asks whether to use the key, and once you approve it, that choice is remembered. In non-interactive mode (-p), the key is always used when it is present. The documentation's precedence list puts ANTHROPIC_API_KEY high in the order Claude Code chooses among credentials — behind cloud-provider credentials and a bearer-token variable, ahead of helper scripts, long-lived OAuth tokens, and stored logins. That ordering is what makes the spare key a real fallback rather than a note in a runbook: when subscription sign-in is the broken component, a session started with the key set does not need the sign-in path at all. It also means the key changes how that session is billed and authenticated, so treat it as a deliberate switch, not a silent one — use it to keep working through the incident, then return to your normal sign-in afterwards.
The audit to run afterwards
The second incident leaves a quieter problem behind. Its 15:33 UTC monitoring update states that scheduled tasks and routines can run more than two hours late and that some runs have been skipped, with Claude Security scans, reviews, and project setup also delayed. A skipped routine does not announce itself in your editor. If a scheduled run produces a report, a summary, or a check you rely on, the absence looks like a quiet day, not a failure. So after an incident day like November 13, audit the routines: list what was scheduled during the affected window, compare it against what actually ran, and re-run anything that was skipped rather than assuming it caught up. The same check covers the late runs — a routine that fired two hours late may have worked on stale inputs or collided with the next scheduled job.
What to do now
Create the spare Console key today, while nothing is on fire, and store it the way you store other credentials. Confirm which role your Console account carries, since that decides whether you can create a key at all. Know how you will start a session with ANTHROPIC_API_KEY set, and know that approving the key once means later sessions remember the choice. Then add one line to your incident routine: when a Claude incident resolves, check the day's scheduled routines for late and skipped runs before you trust their outputs. Neither step helps in the middle of the wider, everything-down incident — nothing on your machine does — but the split shape of November 13, sign-in down and API up, is exactly the shape a prepared user can keep coding through.
Related playbooks
- Keep Claude Code memory under 200 lines
- Claude Code 2.1.318: sign-in expiry and fork effort
- Coding-agent traces in advisory fixes
- GitHub Copilot enters Xcode's coding assistant
Sources
Back to: Coding playbooks hub · Also: Claude Code 2.1.318: sign-in expiry and fork effort · Claude Code 2.1.325 fixes a zstd security issue