
jikkou
Official Jikkou skills for AI agents: manage Kafka topics, ACLs, quotas, schemas, and connectors declaratively with the Jikkou CLI.
- GitHub stars
- 319
Install
/plugin install jikkou@jikkouRun it inside Claude Code after adding the marketplace with /plugin marketplace add streamthoughts/jikkou.
Contents
Discovered from sourceSkills 1
Jikkou: Resource as Code for Apache Kafka
Declarative management of Kafka-ecosystem resources: YAML in, reconciliation against the real cluster out. Stateless: diff compares against live cluster state, never a state file.
Setup check (always run first)
jikkou --versionconfirms the CLI is on PATH. Missing? See "If jikkou is not installed".jikkou health get kafkaverifies connectivity (health get-indicatorsonly lists indicators; it doesn't check connectivity). For other providers usejikkou health get <indicator>; indicator names differ from provider names, seereferences/diagnostics.mdfor the exact list.- No context configured? Run
jikkou config current-contextandjikkou config view; config lives in~/.jikkou/config. See the context safety rule below.
Core workflow
Discover → author → validate → diff → apply. Never skip validate/diff.
- Inspect current state:
jikkou get <provider> <kind> -o JSON(providers: kafka, schemaregistry, kafkaconnect, aiven, confluent-cloud, aws, iceberg). Filter:--name <name>or-s 'metadata.name IN (a,b)'. - Discover kinds/schemas:
jikkou api-resources list, thenjikkou api-resources schema --api-version=<v> --kind=<k>.list → schema → write YAML → validate → diffis the canonical authoring chain. - Validate:
jikkou validate --files <file>returns normalized resources or errors. - Preview:
jikkou diff --files <file>shows the exact changes vs. the live cluster. - Apply:
jikkou apply --files <file> --dry-runfirst; drop--dry-runonly after the user confirms the diff.
Safety rules
- Never run
applywithout first showing the userdiffor--dry-runoutput. - Deleting a resource requires the annotation
jikkou.io/delete: trueon it — never delete another way. - Treat contexts named prod/production with extra confirmation before any non-dry-run apply.
- Never run
jikkou config use-contextcasually: jikkou eagerly loads the current context on every call, so a broken context bricks every command, not just the one you meant to try. Checkcurrent-context/config viewfirst; if a switch is truly needed, verify the target, then switch back.
Provider references
Before authoring resources for a provider, read the matching file:
references/kafka.md, references/schema-registry.md, references/kafka-connect.md, references/aiven.md, references/confluent-cloud.md, references/aws-glue.md, references/iceberg.md.
For misconfiguration checks, consumer lag, and ACL audits, use references/diagnostics.md.
If jikkou is not installed
Confirm with the user first. In order of preference:
- SDKMan:
sdk install jikkou - Homebrew:
brew install streamthoughts/tap/jikkou - No install rights: prefix every command with
docker run --rm -v $PWD:/work -w /work streamthoughts/jikkouAfter install:jikkou --version, thenjikkou health get kafkabefore proceeding.
Common mistakes
- Guessing YAML fields instead of reading the provider reference or running
api-resources schema;validatecatches this, so run it. - Using
kafka-topics.sh-style imperative commands — everything goes through resource files. - Editing a live resource without fetching it first:
jikkou get ... -o YAMLis the starting point. - Applying directly (e.g.
jikkou create -f file.yaml) skippingvalidate/diff/--dry-run, even when the change looks trivial; it mutates the cluster with no preview. - Rediscovering commands via repeated
--helpcalls instead of readingreferences/<provider>.mdfirst (far more tool calls for the same answer). - Switching contexts to "just try" a different cluster — see the context safety rule above.
Sponsors
Details
- Marketplace
- streamthoughts/jikkou
- Category
- Community