Turn failing tests into BugHerd bugs, once each

**Objective:** Log a BugHerd task in the project `{project-name}` for each test in the Ghost Inspector suite `{suite-name}` that failed in the last `{lookback-days}` days (default 1) and has no task yet.

A BugHerd task raised from a failing test carries "ghost-inspector:<test id>" in its external-id, so one test never gets two tasks.

**Step 1: Find the failures**

Call `ghost-inspector-get-suite` with `name.contains` set to `{suite-name}`. If nothing matches, say so and stop. If several match, list their names and ids and ask which one. Then call `ghost-inspector-get-test-result` with `suite-id`, `passing` false and `date-created.from` set to the start of the period. Keep only the latest failing result per test-id, and drop tests whose latest result overall (call `ghost-inspector-get-test-result` with `test-id` and `max-records` 1) is passing again.

**Step 2: Gather the evidence**

For each failing result:

* Call `ghost-inspector-get-test-step` with `test-result-id` and `passing` false: the first failed step's command, target, value and url.
* Call `ghost-inspector-get-console-log` with `test-result-id` and `error` true: up to three error lines.

**Step 3: Skip what is already logged**

Call `bugherd-get-projects` with `name.contains` set to `{project-name}`. If nothing matches, say so and stop. If several match, list their names and project-id values and ask which one; do not guess. For each test, call `bugherd-get-tasks` with `project-id` and `external-id` set to "ghost-inspector:<test id>". If an open task exists, list it under "already logged" and skip the test.

**Step 4: Confirm, then write**

Show the tasks you will create (one per test) and wait for an explicit yes; accept removals. Then for each approved test:

1. Call `bugherd-add-task` once with `project-id`, `external-id` "ghost-inspector:<test id>", `severity` important (critical when the test's start-url is the site's home page or a checkout or login page), and `description`: the test name, the failed step (command, target, value), the step url, the browser and region, the console errors, and the date of the failing run.
2. Call `ghost-inspector-add-test-result-comment` once with `result-id` set to the failing result's id and `message` "Logged in BugHerd as task #<task number>." (read the task number back with `bugherd-get-tasks` and `external-id`).

**Step 5: Report**

Tables of tasks created, tests already logged, and anything that failed.

**Rules & Guidelines:**

* Tools named bugherd-... are on the bugherd MCP server; tools named ghost-inspector-... are on the ghost-inspector MCP server. If either set of tools is missing, say which server is not connected and stop.
* The only write tools this prompt may call are `bugherd-add-task` and `ghost-inspector-add-test-result-comment`, once per approved test.
* Do not re-run tests or accept screenshots.

**Validation:**

* If no test failed in the period, say so and stop.
* A failure on a screenshot comparison only (screenshot-compare-passing false, every step passing) is a visual change, not necessarily a bug: list it separately and do not log it unless the user asks.