What problem does it solve?
Golem agents making HTTP requests, RPC calls, or database operations need controlled retry behavior for transient failures, but hand-rolling backoff logic per call is error-prone. This Skill explains Golem's semantic retry policy system so you can define, apply, and manage retries declaratively.
Core Features & Use Cases
- Manifest-Based Policies: Define named retry policies with priorities and predicates in
golem.yaml under retryPolicyDefaults, using base policies (exponential, periodic, fibonacci, immediate, never) and combinators (countBox, timeBox, clamp, jitter, andThen, union, intersect).
- Runtime SDK Control: Build policies in MoonBit with
Policy and Predicate builders, apply them to scoped code blocks with with_named_policy!, and query active policies via @api.get_retry_policies().
- Live CLI Management: Create, update, list, and delete retry policies at runtime with
golem retry-policy commands without redeploying agents.
- Use Case: Configure an HTTP 5xx retry policy that only fires on transient server errors for HTTPS endpoints, with exponential backoff clamped to [100ms, 5s], 15% jitter, and a maximum of 3 retries.
Quick Start
Ask the AI to add a retry policy to golem.yaml that retries transient HTTPS errors with exponential backoff, jitter, and a maximum of five attempts.
