- name
SKILL.md
스킬 파일은 게시자의 저장소에 그대로 있습니다.
복사한 명령을 터미널에 붙여넣습니다. 이 SKILL.md가 해당 에이전트에 설치됩니다.
npx skills add mariadb-operator/mariadb-operator --skill mariadb-operator-test -a cursor -y복사한 명령을 터미널에 붙여넣습니다. 이 SKILL.md가 해당 에이전트에 설치됩니다.
npx skills add mariadb-operator/mariadb-operator --skill mariadb-operator-test -a claude-code -y복사한 명령을 터미널에 붙여넣습니다. 이 SKILL.md가 해당 에이전트에 설치됩니다.
npx skills add mariadb-operator/mariadb-operator --skill mariadb-operator-test -a codex -y복사한 명령을 터미널에 붙여넣습니다. 이 SKILL.md가 해당 에이전트에 설치됩니다.
npx skills add mariadb-operator/mariadb-operator --skill mariadb-operator-test -a grok -y복사한 메시지를 Grok Bot 채팅에 보냅니다. Bot이 GitHub에서 이 SKILL.md를 가져와 모든 Bot이 공유하는 스킬에 추가합니다.
Install this skill: https://github.com/mariadb-operator/mariadb-operator/tree/HEAD/.agents/skills/mariadb-operator-test게시자 제공
You are a maintainer of mariadb-operator testing a pull request end to end before it is released. The goal is not to confirm that the happy path works — CI already does that — but to break the feature: reproduce the reported problem, inject the failures that happen in real clusters, find the corner cases the author did not think of, fix them, and leave behind the few integration tests that are worth running forever.
Read AGENTS.md at the repository root before doing anything. It describes how to operate the local
environment (§ Development Guides, § Local cluster connectivity, § Local KIND troubleshooting), the testing tiers
(§ Testing), how to reach GitHub (§ GitHub access), the CI gates (§ CI — what a PR must pass), the code patterns
any fix must follow (§ Architecture and Code Patterns) and the guardrails (§ Safety Guardrails). This skill does not
restate them; "§ " refers to a section of AGENTS.md.
mariadb-operator-comment skill and its confirmation gate.waitForIt, transient-state checks), which child objects (Jobs, Pods, PVCs), which SQL state, which
data-plane component (agent, init).Prepare the environment following § Environment setup and § Local cluster connectivity, including the
feature-specific extras the PR needs. Run the operator from the host (make run) so your fixes take effect on every
restart, use debug logging when you need to see why a reconcile is waiting, and stop only the processes you started,
by their exact PID. When something hangs or fails to connect, go through § Local KIND troubleshooting before
suspecting the PR.
Long-running commands (integration suites, slow scenarios) must not exceed a single tool call's timeout: start them detached with a log file and block on the actual test process until it finishes, repeating the wait as needed.
Start from the behavior identified in Step 1 and derive scenarios from these families. Not every family applies to every PR; skip the ones that do not, and say so in the report.
Useful techniques:
--throttle=N in a PhysicalBackup args so mariadb-backup takes minutes).For each scenario: set up, inject, observe, record the evidence, then restore the environment (restart replication threads you stopped, delete the CRs) before the next one.
When a scenario fails:
Keep fixes minimal and in separate, imperative commits that explain the scenario that exposed the gap.
Prove the PR (including your fixes) breaks nothing:
make lint and the unit tests (all suites
must pass).make test-int TEST_ARGS="--focus='…'", then the
broader suites of the affected area (e.g. --focus='PhysicalBackup|MariaDB replication').--json-report=<file> so you can list which specs passed and failed, and why
(State, Failure.Message, Failure.Location); non-verbose Ginkgo output does not name passing specs.origin/main worktree.
A spec that fails the same way on main is pre-existing, not a regression. Report it as such.Write the report and deliver it through the mariadb-operator-comment skill (plain PR comment, confirmation
gate). Keep it about the feature: no reporter's private data, no notes about your local machine. Structure:
## 🧪 End-to-end testing on KIND
<One paragraph: what was tested, operator built from which branch/commit, and the setup (topology, relevant
settings, how the race windows were widened).>
### Scenarios and results
| # | Scenario | Result |
|---|---|---|
| A | <scenario> | ✅ <observed outcome, with the key evidence and timings> |
| B | <scenario> | ❌ → ✅ **Gap, fixed in `<sha>`.** <what broke, what the fix does, the result after it> |
| C | <scenario> | ⚠️ **Not covered**, see below |
### Fixes in `<sha>`
1. **<Fix title>** (`path/to/file.go`): <why it was needed, what changed, why it is safe>.
### Verification
- `make test`: <n> suites pass. `golangci-lint`: <n> issues.
- Integration specs <names>: **<n> Passed, <n> Failed**, <baseline notes for any failure>.
### Open points (pre-existing, not addressed here)
1. **<Problem>.** <Evidence and recommendation, e.g. "needs a finalizer; proposed as a follow-up".>
Use the legend consistently: ✅ passed, ❌ → ✅ failed and fixed in this PR, ⚠️ not covered or needs a decision.
Most scenarios are one-off investigation; only a few deserve to run in CI forever. Evaluate each scenario and propose one or two candidates to the user, with this reasoning:
Ask the user which ones to add before writing them. Then, for each approved test:
internal/controller/*_test.go following the suite's conventions (§ Testing): existing builders and
helpers, Eventually with the suite timeouts, the suite's Ginkgo labels, and an existing Ordered container
when it can reuse its MariaDB (see § Local cluster connectivity for new ones).>
---name: mariadb-operator-testdescription: >---# .agents/skills/mariadb-operator-test/SKILL.mdFAQ
게시자에게서 .agents/skills/mariadb-operator-test/SKILL.md를 다운로드하거나 경로를 복사해 저장소를 엽니다. PlugMCP는 스킬을 목록에 올릴 뿐 파일을 호스팅하지 않습니다.
게시자가 https://github.com/mariadb-operator/mariadb-operator에서 관리합니다.
Claude, Cursor, Codex, Grok Build는 게시자의 저장소에서 추가한 SKILL.md를 읽을 수 있습니다. Grok Bot은 채팅에서 스킬의 GitHub 폴더 링크를 보내면 설치합니다.
| New annotations/fields are missing on existing objects |