rhdh-test-placement
About rhdh-test-placement
An advisory skill for deciding where a lasting RHDH regression or coverage test belongs. It requires `/grilling` first, then recommends a repository, test layer, harness, and existing test to model without writing or running tests.
Key features
- 01Require `/grilling` before placement advice
- 02Recommend repository and test layer
- 03Identify harness and test file to mirror
- 04Compare candidate layers by cost and observability
- 05Leave tests unwritten and files unchanged
Use cases
- 01Choose a layer for a permanent regression test
- 02Find an existing test to use as a template
- 03Decide whether a defect needs a downstream test
Install
Copy the command and run it in a terminal.
npx skills add https:/After installing, type/rhdh-test-placementin Claude Code to confirm it loaded.
Bundled filesagents/openai.yamlreferences/test-placement.md
Details
From the project READMERHDH Test Placement
Advisory only. This skill answers where a lasting test belongs and what it should be modelled on. It does not write the test, run it, or change a file. The guiding rule: pick the cheapest environment that can actually catch the bug. Most plugin validation needs no cluster, and a growing share needs no Docker either.
Prerequisite: /grilling
Placement is a tradeoff, not a lookup. Cost, feedback time, coverage overlap, and what a layer can physically observe all pull in different directions, and the person asking usually has a preferred answer already.
Invoke /grilling before recommending a placement, and use the constraints
it produces. If /grilling is not installed, stop: say the skill is required,
name /setup-rhdh-skills install as the human's next step, and do not guess a
placement in the meantime.
Scope
| In scope | Out of scope |
|---|---|
| Where a permanent regression or coverage test belongs | Writing or running the test |
| Which repository and layer, and why not the others | Fixing the defect under test |
| Which existing file to mirror | A throwaway reproduction test |
| The rough feedback cost of each candidate | Reviewing an RHDH Jira test plan |
A one-off test that reproduces a defect and is then deleted is not this skill's
subject; /rhdh-plugin-bug-fix owns that and removes the file before handing
its change off. Come here for the durable test that stays behind afterwards.
Route by outcome
| Outcome | Load and follow |
|---|---|
| Recommend a repository, layer, harness, and template | references/test-placement.md |
references/test-placement.md carries the context questions to settle first,
the repository decision table, the L1–L4b layer ladder, per-placement authoring
notes with the file to copy, the commands that run each layer, and the things
that are researched as not possible today.
Boundaries
/grillingis a hard prerequisite, not an option./rhdh-plugin-authoringowns writing the recommended test once the placement is settled./rhdh-plugin-bug-fixowns the throwaway reproduction and its evidence./rhdh-test-plan-reviewreviews an RHDH release test plan in Jira. That is a release-readiness review, not a placement question.
Invoke a named skill and describe the handoff in the conversation. Never open another skill's files.
Invariants
- Verify every referenced path against the target repository's current default branch before recommending it. Harnesses in this domain move, and some cited here were still in review when they were written down.
- Justify the test by the failure it would catch, never by a coverage number.
Codecov is informational on both
rhdhandrhdh-plugins; no coverage delta can block a pull request, and a test argued that way gets rejected in review. - If a defect is catchable at more than one layer, recommend the cheapest and say explicitly not to duplicate it downstream.
- Never recommend a UI-render test in
rhdh-plugin-export-overlays; that repository has no app to render into. - Say "no good placement exists" when that is the answer. Recommending an expensive layer that still cannot observe the failure is worse than none.
Completion
Report the repository, the layer and harness, the directory or file the test goes in, the existing neighbour to mirror, one line on each layer rejected and why, and the rough feedback cost. Leave the test unwritten and no file changed.