Investigate a failing Ghost Inspector test with its Jira history

**Objective:** Explain why test {test-name} is failing, by combining its Ghost Inspector run history with everything Jira project {jira-project-key} already records about it, over the last {lookback-days} days (default 30). This prompt reports only. It creates and changes nothing.

**Step 1: The test itself**

Call `ghost-inspector-get-test` on the ghost-inspector server with `name.contains` set to {test-name}. Keep its id, suite, browser, region, start URL, frequency, screenshot comparison setting and auto retry setting. If more than one test matches, list them and ask which to investigate.

**Step 2: The run history**

Call `ghost-inspector-get-test-result` with `test-id` set to the test's id and `date-created.from` set to the start of the window, `max-records` 100.

Work out the pass rate, whether failures cluster on one browser or region, and whether they cluster in time. A test that fails on one browser only, or fails in bursts, is a different problem from one that fails steadily, and the report must say which of the three this is.

**Step 3: What the last failure actually did**

Take the most recent failing result. Call `ghost-inspector-get-test-step` with `test-result-id` set to its id and `passing` set to false, and keep the first failing step in sequence order. Call `ghost-inspector-get-console-log` with the same `test-result-id` and `error` set to true.

Call `ghost-inspector-get-test-result-comment` with the same `test-result-id` to pick up anything a person already noted.

**Step 4: What Jira already knows**

Call `jira-get-issue` on the jira server with `project-key` set to {jira-project-key} and `summary.contains` set to {test-name}, `max-records` 50. Keep each issue's key, summary, status, assignee, creation date and updated date.

For the most recently updated match, call `jira-get-issue-comment` with `issue-id` set to that issue key and read the comment history. This is where a previous investigation's conclusion lives, and repeating work already done is the failure mode this step exists to prevent.

If the failing step's target names a page or component, call `jira-get-issue` again with `summary.contains` set to that term to find issues about the underlying feature rather than the test.

**Step 5: Report**

Produce, in order: what the test does and where it runs; the pass rate and the failure pattern, named as browser-specific, time-clustered or steady; the failing step and the console errors; every Jira issue that mentions this test, with key, status and the one-line conclusion from its comments; any related issue about the underlying feature; and finally your own assessment of whether this is a product bug, a test that needs updating, or environmental flakiness, with the evidence for it.

**Rules & Guidelines:**

* `ghost-inspector-get-test`, `ghost-inspector-get-test-result`, `ghost-inspector-get-test-step`, `ghost-inspector-get-console-log` and `ghost-inspector-get-test-result-comment` are on the ghost-inspector server. `jira-get-issue` and `jira-get-issue-comment` are on the jira server. All three servers must be connected.
* The linking convention used by every prompt on this server: a Jira issue raised from a Ghost Inspector failure has a summary beginning `Ghost Inspector failure: ` followed by the test name, and its first comment contains the line `Ghost Inspector result id: ` followed by the result id. Matching in both directions depends on this, so do not vary the wording.
* Do NOT call any write tool. Specifically do not call `jira-add-issue`, `jira-add-issue-comment`, `jira-update-issue`, `ghost-inspector-execute-test`, `ghost-inspector-add-test-result-comment`, `ghost-inspector-update-test` or `ghost-inspector-accept-screenshot`.
* Do not assert a cause you cannot point at a step, a console error or a ticket for. Say the evidence is inconclusive instead.

**Validation:**

* If the test has no failing runs in the window, report it as currently healthy and give the pass rate anyway.
* If Jira returns no issue mentioning the test, say the failure is untracked, and note that the raising prompt on this server can file it.
* If Ghost Inspector and Jira disagree, for example an issue is Done while the test still fails daily, call that out at the top of the report rather than burying it.