Redpanda Agent Skills
Agent Skills for Redpanda's five products — Streaming (the Kafka-compatible engine),
SQL, Connect, Cloud (Serverless, BYOC, Dedicated), and the Agentic
Data Plane — unified by the rpk command-line experience. Each skill is grounded
in Redpanda's own source code, documentation, and APIs.
Installation
One command (from your shell):
claude plugin marketplace add redpanda-data/skills && claude plugin install redpanda@redpanda-skills
Or inside a Claude Code session, as two separate slash commands (run one at a time —
don't paste them together):
/plugin marketplace add redpanda-data/skills
/plugin install redpanda@redpanda-skills
Once installed, invoke any skill with its /redpanda:<skill> name (e.g.
/redpanda:streaming). Each skill's SKILL.md frontmatter description tells the agent
when to load it automatically.
Keep the skills up to date
The skills change as the products do. Claude Code doesn't auto-update plugins from this
marketplace unless you turn it on: run /plugin, open the Marketplaces tab, select
redpanda-skills, and select Enable auto-update. To update by hand:
claude plugin marketplace update redpanda-skills && claude plugin update redpanda@redpanda-skills
What is this?
This repo is a Claude Code plugin named redpanda. When installed, every skill is
available to the agent as /redpanda:<skill> (for example /redpanda:streaming).
A skill is a focused, on-demand reference: a SKILL.md the agent loads when relevant,
plus references/ files it pulls in for depth. Skills are self-contained and contain
copy-pasteable commands, configs, and code.
Available Skills
Redpanda has five public products — Streaming, SQL, Connect, Cloud, and the Agentic
Data Plane — with rpk as the unifying CLI across all of them.
🌊 Streaming
The Kafka-compatible streaming engine (the Redpanda broker).
- streaming — The Kafka API: produce/consume, consumer groups, transactions & idempotence, topic management, client compatibility, tiered storage, Iceberg topics, cloud topics, continuous balancing, and shadow linking.
- streaming-admin-api — Operate the cluster via the HTTP Admin API (port 9644): brokers, partitions, cluster config, features, licensing, transactions, debug, and debug bundles.
- streaming-debugging — Diagnose a broker/cluster: debug bundles, metrics endpoints, logs, CPU profiling, partition/raft health, and failure-mode playbooks.
🧮 SQL
Redpanda SQL — the distributed, columnar, PostgreSQL-compatible query engine on Redpanda Cloud BYOC clusters.
- sql — Write SQL: connecting and authenticating, data types, read-only queries, functions, and analytical queries.
- sql-admin-api — Administer Redpanda SQL on BYOC: enable, scale, and disable it through the Cloud Console, Cloud API, or Terraform, plus authentication and the access model.
- sql-federated-queries — Query Redpanda topics through catalogs, including Iceberg-enabled topics and nested fields.
- sql-debugging — Diagnose: system virtual tables, degraded nodes, query out-of-memory errors, slow queries, and when to contact Redpanda Support.
🔌 Connect
Redpanda Connect — declarative stream processing (formerly Benthos).
- connect — Build pipelines (inputs, processors, outputs, Bloblang), the config structure, and running pipelines.
- connect-debugging — Debug pipelines: linting, log levels, metrics, tracing, health endpoints, dry-runs, and common failure modes.
Change Data Capture (one per connector):
☁️ Cloud
Redpanda Cloud — the managed control plane and per-cluster data plane.
- cloud-serverless — Serverless clusters (multi-tenant, pay-per-use) via the Control Plane API, plus the per-cluster Data Plane API.
- cloud-byoc — BYOC (Bring Your Own Cloud) clusters in your own AWS/GCP/Azure account: networks/VPCs, provider setup, and the
rpk BYOC agent flow.
- cloud-dedicated — Dedicated clusters (fully Redpanda-managed, single-tenant, in Redpanda's cloud account) via the Control Plane API.
🤖 Agentic Data Plane
Redpanda's governance infrastructure for AI agents and MCP servers — its own product surface that runs on Redpanda, provisioning its own environment.
- adp — Build and operate AI on Redpanda: AI agents, MCP servers (managed catalog + code mode), the AI Gateway (LLM providers and models), governance (budgets, guardrails, Cedar policies, OAuth), and observability (transcripts) — driven through the
rpk ai CLI and the Agentic Data Plane API.
⌨️ rpk — the unifying CLI
One CLI across every product.
- rpk — Install rpk, configure connections with profiles and
-X flags, and target self-hosted or Redpanda Cloud clusters. Start here.
- rpk-topic — Create, describe, alter, and delete topics; produce and consume records.
- rpk-cluster — Health/metadata, brokers, cluster config, partition balancing/movement, maintenance mode, quotas, storage (mountable topics / whole-cluster restore), and self-tests.
- rpk-redpanda — Operate a self-managed broker node: start/stop the process, production/dev/recovery mode, the autotuner and
rpk iotune, node config bootstrap, and broker decommission/recommission via the admin listener. Self-managed only.
- rpk-group — List/describe consumer groups, inspect lag and members, reset/seek offsets, and delete groups or offsets.
- rpk-security — Manage SASL/SCRAM users, Kafka ACLs, RBAC roles, and secrets.
- rpk-cloud — Authenticate to Redpanda Cloud, manage credentials, select clusters, manage resource groups, drive BYOC provisioning, and run the
rpk cloud mcp server that exposes Redpanda Cloud management to AI agents.
- rpk-debug — Collect local and remote debug bundles and inspect local process info.
- rpk-registry — Manage Schema Registry subjects, schemas, compatibility, and modes.
How these skills were built
Each skill was drafted from a grounded spec (source paths into the Redpanda, Connect,
and Redpanda SQL source plus the docs), then put through an adversarial review at maximum
model effort that cross-checks every command, flag, config field, endpoint, and code
example against the actual source — fixing hallucinated or outdated details — followed by
a dedicated enterprise-feature coverage pass (the key differentiators, with their
nested settings) and a final verification pass.
How these skills stay current
The skills are kept in sync with Redpanda's source by a fleet of scheduled maintenance
routines — a generator → critic → human-merge loop per product, plus a monthly drift
audit — covering all five products. See MAINTAINING.md for how the
process works and how to operate it. Each source-grounded skill carries a
references/SOURCES.md that maps its claims to the upstream source.
License
Distributed under the Apache 2.0 license. See the LICENSE file.
Acknowledgements
Structure and format modeled on google/skills.
