kurrent

Use when asked to review KurrentDB client code, after writing or modifying SDK call sites, as the post-migration check dispatched by migration-specialist, or as a pre-PR gate. Reviews application code that uses the KurrentDB / EventStoreDB client SDKs for idiomatic usage, anti-patterns, and (in post-migration mode) migration completeness. Flags stream-naming mistakes, lost optimistic-concurrency guards, per-call client instantiation, missing retry pipelines, ad-hoc subscription wiring, mishandled SDK exceptions, and stale connection strings. In post-migration mode also flags leftover legacy types, silent concurrency downgrades against git history, and failing builds. Do not use for runtime smoke flows (point the user at the troubleshooter agent or run the project's own integration tests instead).

Use when asked to migrate off the TCP client, retire EventStoreConnection, move from esdb:// to kurrentdb://, or upgrade an intermediate gRPC client to KurrentDB.Client. Orchestrates KurrentDB / EventStoreDB client SDK migrations: detects the current client and connection scheme, routes to the matching flavour of the kurrent-upgrade skill (tcp-to-grpc rewrite or EventStoreDB gRPC rebrand), and walks the user through the migration.

Use when diagnosing KurrentDB / EventStoreDB runtime failures: connection errors, TLS handshake failures, WrongExpectedVersion conflicts, subscription drops or lag, persistent-subscription parking, cluster leader-election issues, gossip partitions, scavenge hangs, projection divergence. Walks decision trees per failure mode, dispatches kurrentdb-client-detection and kurrentdb-server-detection to gather state, and produces a graded report with cause, fix, and verification step. Typical triggers include "my app can't connect to my cluster", "subscription is lagging", "WrongExpectedVersion in production", "cluster won't elect a leader", "scavenge is stuck". See "When to invoke" in the agent body for worked scenarios. Do not use for static code review (use code-reviewer) or migration walkthroughs (use migration-specialist).

Use when migrating an app onto the KurrentDB gRPC client from a legacy EventStoreDB client. Covers both the TCP rewrite and gRPC rebrand paths, plus the esdb to kurrentdb connection-string switch. NOT for code already on the KurrentDB client (use kurrent-docs) or server upgrades.

Use when inventorying an application's KurrentDB or EventStoreDB client surface before a migration or audit. NOT for server or cluster state (use kurrentdb-server-detection).

Use when working on, updating, or reviewing code that instantiates or configures a KurrentDB gRPC client across the supported SDKs. Covers connection-string parameters, single-node versus gossip discovery for clusters, node preference for read and write splits, HTTP/2 keepalive for long-lived connections behind NAT or load balancers, default deadlines, serverless client reuse, and triage of connection-related failures like discovery timeouts, deadline exceeded, keep-alive resets, and TLS verification errors. Do NOT use for authentication mechanics, certificate provisioning on the server, EventStoreDB to KurrentDB migration, or server-side cluster sizing.

Use when inventorying a deployed KurrentDB or EventStoreDB server for its version, cluster topology, license status, and deployment method by reading runtime endpoints, deployment manifests, and on-disk version banners. NOT for client SDK state (use kurrentdb-client-detection), Kurrent Cloud projects, or migration execution.

Use when the user is working with KurrentDB, from writing application code against the client SDKs to operating the self-hosted server or running on Kurrent Cloud.