跳至正文
OpenAgentSkill
浏览已开放 · 目录记录日期 2026-11-21 · 自动更新和账户服务暂时关闭。
数据与分析

create-kit

Create a new easy-db-lab kit (kit.yaml + K8s manifests, metrics, dashboards, docs) for any database or workload. Detects internal (the easy-db-lab repo) vs external (your own project) mode, researches the workload, grounds the work in a per-kit TODO.md, drafts for review, and writes deliverable files only on explicit approval. Use when adding a new database or workload kit to easy-db-lab or to your own kit source.

rustyrazorblade · 已有静态检查记录

安装与使用

运行命令前先阅读来源与依赖说明。复制命令不会自动安装或执行。

npx skills add rustyrazorblade/skills --skill create-kit

技能说明


name: create-kit description: Create a new easy-db-lab kit (kit.yaml + K8s manifests, metrics, dashboards, docs) for any database or workload. Detects internal (the easy-db-lab repo) vs external (your own project) mode, researches the workload, grounds the work in a per-kit TODO.md, drafts for review, and writes deliverable files only on explicit approval. Use when adding a new database or workload kit to easy-db-lab or to your own kit source. argument-hint: "[workload name] [output path]" user-invocable: true

create-kit

Author a new easy-db-lab kit — a self-contained package (kit.yaml + manifest templates + dashboards + docs) that installs, starts, stops, and observes a database or workload on an easy-db-lab cluster. No Kotlin is required; kits register dynamically from their kit.yaml.

This skill works for both:

  • Internal kit development — inside the easy-db-lab repository (src/main/resources/.../kits/).
  • External kit development — a kit that lives in your own project and is registered with easy-db-lab kit source add.

The skill detects which mode applies and adapts paths and the schema source accordingly.

Input

The argument after /create-kit is the workload name (e.g. "ScyllaDB", "Kafka", "Redis"). An optional second argument is the output path. If no workload name is given, ask for one.


Step 0 — Detect mode and resolve paths

Check whether the current directory is the easy-db-lab repository:

test -d src/main/resources/com/rustyrazorblade/easydblab/kits && echo INTERNAL || echo EXTERNAL

Internal mode (the path exists):

  • Kit directory: src/main/resources/com/rustyrazorblade/easydblab/kits/<name>/
  • Authoritative schema source: read the local files docs/development/kits.md and src/main/kotlin/com/rustyrazorblade/easydblab/services/KitConfig.kt. They are version-matched to this checkout — trust them over the cheat-sheet below.

External mode (the path does not exist):

  • Kit directory: ./kits/<name>/ by default. If the user passed an output path, use it. The kit's parent directory (./kits) is what gets registered with kit source add.
  • Authoritative schema source: WebFetch the published guide https://rustyrazorblade.github.io/easy-db-lab/development/kits.html whenever you need anything beyond the cheat-sheet below. (Repo: https://github.com/rustyrazorblade/easy-db-lab/.)

<name> is the kebab-case kit name. State the detected mode and resolved kit directory to the user before proceeding.


Step 1 — Research the workload

Use WebSearch and/or context7 to determine:

  • The canonical Helm chart (repo URL, chart name, current stable version), or an operator/CRD that must be installed first (operator chart → then the custom resource).
  • The primary Kubernetes resource kind the workload creates (StatefulSet, Deployment, or a custom CR) and the readiness condition to wait on (pod labels, conditions).
  • Whether the workload is stateful (needs platform-pvs for local storage) or stateless.
  • Metrics: does it expose a Prometheus endpoint? Port and path? Or is it JVM-based (OTel/Pyroscope java agent), or does the chart ship its own OTLP pipeline (helm-native)?
  • Endpoints / wire protocols: what ports clients connect on (HTTP, JDBC, native, CQL, PostgreSQL/MySQL wire), and whether it exposes a SQL interface (declare the sql capability so bench kits like sysbench can target it).
  • JVM?: if yes, plan Pyroscope java-agent injection in the start phase.
  • Integration with existing kits: does this workload need a backing store or data source another kit provides (e.g. Temporal needs PostgreSQL/Cassandra; Trino/Presto register catalogs for ClickHouse/Cassandra/TiDB)? Or is it a bench/consumer kit that targets a running database? Check easy-db-lab kit list / the kits dir for what already exists, then pick a wiring mechanism (see Integrating with other kits below).

Summarize findings in 3–5 bullets — including the recommended metrics.type, whether Pyroscope injection is needed, and any cross-kit integration — before grounding the work.

Integrating with other kits

Three mechanisms wire kits together — pick by the relationship:

  • kit-ref arg (--target) — a kit that targets a user-chosen running kit (bench/consumer kits). Declare { flag: --target, variable: TARGET, type: kit-ref, capability: sql }. At start, KitEndpointResolver reads the target kit's endpoints and injects TARGET_* env vars (TARGET_PG_HOST/PORT/USER/DATABASE, TARGET_JDBC_URL, TARGET_MYSQL_HOST, …) into every phase; the installed dir becomes <kit>-<target>. Reference: sysbench.
  • Hooks (post-workload-start / post-workload-stop) — a kit that reacts when another kit starts or stops (e.g. registering a connector/catalog). Optionally scope with workloads: [postgres, clickhouse] (the YAML field is workloads, not kits); omit to fire for any kit. Reference: Trino's bin/update-catalogs.sh, which re-registers catalogs after ClickHouse/Cassandra/TiDB start.
  • In-cluster service DNS — a hard dependency on another kit's service: reference its stable cluster DNS directly in a manifest/values template (e.g. clickhouse-clickhouse.default.svc.cluster.local:8123). Use when the backing kit is assumed present.

For a SQL database others should be able to target, declare the sql capability so kit-ref consumers can find it.

Reading another kit's values into a configmap or manifest

Typed-step string fields are substituted with the ${VAR} pattern against the runtime variable set (the same set shell: steps receive as environment variables). This includes configmap data: values, label values, exec command/namespace/pod, and manifest bodies when the step sets interpolate: true. So to write another kit's connection info into a configmap (e.g. you need THE_POSTGRES_ENDPOINT for your own reas

阅读完整来源 ↗