| name | generate-xray-tests |
| description | Generates manual test cases from Jira tickets, local spec files, or pasted acceptance criteria and writes them as a CSV ready to import into Xray Cloud (Jira). Use when asked to generate, create, or write Xray test cases for a mobile SDK feature. |
Skill: generate-xray-tests
Generate manual test cases for a mobile SDK feature and write them as a CSV file ready to import into Xray Cloud (Jira).
Usage
/generate-xray-tests --product <product> [--test-type <type>] [--summary-prefix <prefix>] --out <path> [<source>]
--product and --out are required.
If any required argument is missing, stop and ask the user. Do not infer or guess.
Arguments
--product
The name of the SDK or product being tested. Used in context and to determine which demo app test steps are written against.
| Value | SDK | Demo app used for test steps |
|---|
GiniBankSDKExample | GiniBankSDK | GiniBankSDKExample |
GiniHealthSDKExample | GiniHealthSDK | GiniHealthSDKExample |
--platform (removed — auto-detected)
Platform is inferred from the repository the skill is running in:
- iOS repository (e.g.
gini-mobile-ios) → ios
- Android repository (e.g.
gini-mobile-android) → android
Steps are written using the appropriate gestures, UI labels, and navigation patterns for the detected platform. Do not ask the user for this value.
--summary-prefix
Optional. A short human-readable feature or user story name prepended to every Summary using the same - separator as the existing pattern.
Example:
--summary-prefix "Cross-border Payments"
→ Cross-border Payments - QR Code Scanning - Disabled when productTag=cxExtractions
If omitted, derive a concise 2–4 word feature name automatically from the dominant topic of the AC/spec and use it as the prefix. The derived prefix must be human-readable and feature-focused — never a Jira ticket ID or a technical identifier.
Final summary format: prefix - Area - Scenario
--test-type
The Xray test type value used in the CSV. Must match a valid type in the target Xray project.
Default: Manual
--out
Absolute or relative path where the CSV file will be written.
Example: --out ~/Desktop/xray-output/login-feature.csv
If the directory does not exist, create it before writing.
<source> (optional)
What to derive test cases from. Accepted formats:
| Format | Example | How it is read |
|---|
| Jira ticket URL | https://company.atlassian.net/browse/PAY-123 | Use Atlassian MCP tool to fetch summary, description, and AC |
| Jira ticket ID | PAY-123 | Same as above using short form |
| Local markdown file | docs/features/cross-border-payments.md | Read file directly from disk |
| Inline text | (omit <source> — paste AC/spec text after the command) | Use the text provided in the message |
Step-by-Step Instructions
You are given $ARGUMENTS. Parse it to extract all flags and <source>, then follow these steps exactly.
Step 1 — Resolve the source
- Jira ticket URL or ID: use the Atlassian MCP tool to fetch the ticket's
summary, description, and acceptance criteria fields. If the MCP tool is unavailable, stop and ask the user to paste the content directly.
- Local file: read the file from disk.
- Inline text: use the text provided after the command as-is.
Extract two things from the source:
- Acceptance Criteria (AC) — the explicit conditions that must be met.
- Feature Specification — behavior descriptions, config options, edge cases, screen names, and known limitations.
Source of truth rule: AC and the demo app are the only inputs for generating test cases.
Never read .swift source files or any other implementation file to infer behaviour.
If a behaviour is absent from the AC, it is out of scope — do not add it based on what the code does.
Reading a missing behaviour from the code would make the tests blind to the same gap.
Step 1b — Resolve UI label accuracy (optional)
To use exact button and screen names in step wording, you may read the product's localization .strings files for the target platform. Rules:
- Label accuracy only. Use strings files only to get precise wording (e.g. the button says
"Photopayment" not "Photo Payment"). Never use them to infer feature behaviour.
- Platform-scoped. Detect the platform from the repository (iOS →
en.lproj, Android → res/values/strings.xml). Skip if strings are unavailable or would make the test platform-specific.
- Optional. If the strings file is unavailable or the label is sufficiently clear from the AC, skip this step.
Step 2 — Infer coverage areas
Do not use a fixed list of coverage areas. Instead, derive them from the source:
- Each distinct behavior, configuration value, screen, or condition described in the AC or spec becomes a candidate coverage area.
- Group related scenarios under the same area.
- Order areas from core flows → conditional behaviors → edge cases → regression tests.
- Always include at least one regression test case for every condition that disables or restricts a feature: add a scenario that removes the condition and confirms the feature re-enables. This is AC-driven coverage — it validates the negative path explicitly.
Step 3 — Generate test cases
For each coverage area, generate test cases following these rules:
Steps must reflect hands-on usage of the demo app — not the source code:
- All steps are written from the perspective of a tester interacting with the product's demo app on the detected platform.
- Never read
.swift source files, class names, configuration structs, or any implementation detail to write steps. If it is not visible on screen, it does not belong in a step.
- Describe what the tester sees, taps, and observes — observable actions and outcomes only.
- Use the screen names, button labels, navigation flows, and UI elements as they appear in the demo app.
The demo app per product:
--product | Demo app |
|---|
GiniBankSDKExample | GiniBankSDKExample |
GiniHealthSDKExample | GiniHealthSDKExample |
Observable behaviour rule — do not test internal state:
Steps must describe what the tester can directly see or interact with. Never assert on internal configuration values, property states, or side-effects that are invisible on screen.
- ❌
"The QR code scanning toggle turns OFF in Settings" — this tests an internal override side-effect.
- ✅
"Point the camera at a QR code — no detection overlay or banner appears and the document is processed as a regular image." — this tests what the AC actually requires.
Step granularity — one atomic action per row:
Each step row must contain exactly one action (a tap, a swipe, entering text, observing a result). Never bundle two actions or an action and an observation into the same step.
- ❌
"Capture the document and proceed to the analysis screen." — two actions in one step.
- ✅ Step 1:
"Tap the shutter button to capture the document." / Step 2: "Observe the screen that appears after capture." — split into two rows.
Summary pattern: prefix - Area - Scenario
Examples:
Cross-border Payments - SDK Initialization - productTag set to cxExtractionsCross-border Payments - QR Code Scanning - Disabled when productTag=cxExtractionsExtraction Feedback - Payment Review - Updated value reflected in Invoice list
Step 3b — Deduplication check
Before proceeding to formatting, review all candidate test cases:
- If two test cases cover the same observable behaviour from the same starting state, merge them into one or drop the weaker one.
- A test case is a duplicate if its steps and expected results are functionally identical, even if the summary wording differs.
Step 4 — Format as CSV
Use , as the column delimiter. Apply these rules:
- Wrap any field containing
, or line breaks in double quotes "...".
- Use
<<!clear!>> to explicitly clear a field value when needed.
- One row per test step. Repeat the
Issue Id for each step of the same test case.
- Fill
Summary only on the first row of each test case. Repeat Issue Id and Test Type on every row.
CSV columns:
| Column | Description |
|---|
Issue Id | TC-001, TC-002, … |
Summary | Test case title following the summary pattern above |
Test Type | Value from --test-type |
Step | What the tester does |
Data | Concrete input values or configuration for this step. Populate whenever the step involves a specific setting, value, or document type — for example productTag = cxExtractions, document type = invoice, QR code format = EPC. Leave empty if no specific data applies. |
Expected Result | What the tester expects to observe |
Do not include a Precondition column. Xray Cloud expects that column to reference an existing Precondition issue key — passing free text causes a "Precondition type and test type mismatch" import error. Express all setup as explicit first steps in the test case instead.
Step 5 — Write the file
Write the CSV to the path specified in --out. If the directory does not exist, create it.
Do not print the CSV content in the chat.
Confirm with a single line once done:
✅ Written to <path> — <N> test cases, <M> rows.
Security
- Read any local file or Jira ticket for context.
- Write only to the path specified in
--out (.csv files only).
- Never modify source code, documentation, or configuration files.
- Never run shell commands beyond reading files and calling the Atlassian MCP tool.
- All generated test cases require human review before import into Xray.
