Anthropic has implemented a built-in feature in Claude Code for automatically resuming sessions after usage quotas are exhausted. The desktop version received a toggle, and the CLI received an environment variable that configures retry logic up to 300 retries and enables infinite retry attempts for 429 and 529 errors. Limits are not increased in this case — the feature simply removes the need to manually run continue after each reset.

image
image

What happened

The desktop version of Claude Code now has a “Continue after reset” checkbox. When enabled, a session stopped due to a 5-hour or weekly limit will automatically resume as soon as the quota is updated. In the CLI version, a similar feature is activated by the environment variable CLAUDE_CODE_RETRY_WATCHDOG=1. This flag increases the number of retries to 300 and allows infinite retries for HTTP errors 429 and 529. The feature is documented in the Environment variables and Error reference sections of the official Claude Code documentation.

Context

On August 14, Anthropic made Auto Mode the default setting for Pro, Max, and Team plans — agents began performing actions without explicit confirmation at each step. Auto-continuation of sessions logically continues this same strategy: minimizing manual intervention not only when approving actions but also when interrupted due to limits. Anthropic’s approach to UX automation with fixed quotas contrasts with the methodology of competitors such as OpenAI Codex, which, in response to negative user reactions, applies blunt quota resets for the entire user base.

Why this matters for the industry

The release confirms Anthropic’s strategic shift toward fully autonomous agent workflows. The retry-after-limit pattern with standard exponential backoff logic, packaged in a user interface, sets a benchmark for competing agent platforms: Codex, Gemini CLI, Cursor. Similar retry mechanisms and third-party orchestration tools for multi-provider pipelines are expected to appear. In the long term, the auto-resume pattern after a rate limit will become an expected baseline for any agent platform, and eval benchmarks may shift from single-session metrics to multi-session and multi-day task completion rates.

Why this matters for users

Users who run long Claude Code sessions with /goal, /schedule commands, or background tasks will no longer be permanently interrupted after the 5-hour window. In the desktop version, it is enough to check one box; in the CLI, set CLAUDE_CODE_RETRY_WATCHDOG=1. One flag replaces wrapper scripts that developers wrote manually for retry logic. The feature does not increase limits and does not solve the problem of weekly quota shortage — if the main block is the volume of the weekly quota, auto-continuation will not eliminate it.

What is still unknown / limitations

The feature is a standard retry loop with exponential backoff, packaged in an interface — this is an infrastructure pattern that has been used in engineering for decades, not a scientific or model breakthrough. It is unknown whether Anthropic plans to scale the retry logic to other products. Infinite retries create implicit SLA risks for tasks with long waits for quota resets.

Sources

Author

Look at AI, editorial team