Skip to main content
When a task fails, Claude Board automatically retries it with increasing delays to avoid hammering rate limits and give transient issues time to resolve.

​
How It Works

Failed tasks follow an exponential backoff strategy with jitter:
RetryBase DelayWith Jitter (±20%)
1st30s24–36s
2nd60s48–72s
3rd120s96–144s
4th240s192–288s
5th+600s (max)480–600s

​
Retry Flow

Task fails

The Claude process exits with a non-zero exit code (crash, rate limit, timeout, etc.)

Retry check

Claude Board checks if retry_count < max_retries. If yes, the task is moved back to Backlog with a retry_after timestamp.

Backoff delay

The queue poller skips the task until the retry_after timestamp expires.

Auto-restart

Once the delay expires, the queue picks up the task and starts a new agent.
If all retries are exhausted, the task moves to Failed status permanently.

​
Configuration

Set max retries per project in Project Settings > Automation:
SettingDefaultRangeDescription
Max Retries20–10Number of retry attempts before permanent failure
Set max retries to 0 to disable automatic retrying. Failed tasks will immediately move to Failed status.

​
Manual Reset

Moving a failed task back to Backlog or In Progress resets the retry counter to 0, giving it a fresh start.

​
When Retries Trigger

Retries activate when a task fails due to:
  • Rate limiting — Claude API returns 429
  • Process crash — unexpected exit code
  • Timeout — task exceeds the configured timeout
  • Auto-test rejection — tests fail and auto-revision is enabled
Tasks that complete successfully but fail during auto-testing follow a separate auto-revision flow before entering the retry cycle.