Managing kafka resources

Skill streamthoughts/jikkou/skills/managing-kafka-resources

Use when managing Apache Kafka topics, ACLs, quotas, users, consumer groups, Schema Registry subjects, Kafka Connect connectors, or Aiven, Confluent Cloud, AWS Glue, and Apache Iceberg resources declaratively with the Jikkou CLI: creating, inspecting, or changing resources, previewing changes before apply, or diagnosing Kafka misconfigurations.From its SKILL.md

Install

npx -y skills add streamthoughts/jikkou --skill managing-kafka-resources

Assembled from the repository path, not quoted from the project. Check it against their README if it does not work.

SKILL.md

4.0 KB, as published. Nobody here has run it

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 --version confirms the CLI is on PATH. Missing? See "If jikkou is not installed".
jikkou health get kafka verifies connectivity (health get-indicators only lists indicators; it doesn't check connectivity). For other providers use jikkou health get <indicator>; indicator names differ from provider names, see references/diagnostics.md for the exact list.
No context configured? Run jikkou config current-context and jikkou 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, then jikkou api-resources schema --api-version=<v> --kind=<k>. list → schema → write YAML → validate → diff is 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-run first; drop --dry-run only after the user confirms the diff.

Safety rules

Never run apply without first showing the user diff or --dry-run output.
Deleting a resource requires the annotation jikkou.io/delete: true on it — never delete another way.
Treat contexts named prod/production with extra confirmation before any non-dry-run apply.
Never run jikkou config use-context casually: jikkou eagerly loads the current context on every call, so a broken context bricks every command, not just the one you meant to try. Check current-context/config view first; 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/jikkou After install: jikkou --version, then jikkou health get kafka before proceeding.

Common mistakes

Guessing YAML fields instead of reading the provider reference or running api-resources schema; validate catches 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 YAML is the starting point.
Applying directly (e.g. jikkou create -f file.yaml) skipping validate/diff/--dry-run, even when the change looks trivial; it mutates the cluster with no preview.
Rediscovering commands via repeated --help calls instead of reading references/<provider>.md first (far more tool calls for the same answer).
Switching contexts to "just try" a different cluster — see the context safety rule above.