Layering QA Skills Before AI Automation: Lessons from the Agent Era

Why QA Needs Skill Layering in the AI Agent Era

AI can now write code faster than ever, but the pressure in the development pipeline does not disappear; it moves toward verification and delivery. Issues such as misunderstood requirements, missed test points, core‑path failures, incorrect interface results, unstable real‑device regressions, and incomplete release packages increasingly fall on QA.

To address this, we reorganized the testing process. Tasks that require business experience and risk judgment remain with QA, while repetitive, rule‑based steps with well‑defined inputs and outputs are distilled into callable, inspectable QA Skills and platform capabilities.

01 Why QA Should Layer Skills Along the Test Process

Repeated work appears not only when writing test cases but also during self‑testing, smoke testing, data preparation, cache checks, real‑device regression, and pre‑release packaging. Although each action exists in the traditional workflow, they are repeatedly performed across different requirements and versions, leading to quality dependence on personal experience, ad‑hoc communication, and manual checks.

Our approach extracts standardizable steps layer by layer, ensuring each capability can be invoked independently, its execution can be inspected, and critical operations still require human confirmation. Once these single‑point abilities stabilize, we address context passing, result hand‑off, and process chaining.

02 Automation Executes Repetitive Tasks; QA Makes the Judgment

Decomposing the test flow into Skills does not hand over quality judgment to AI. Skills are suited for generation, query, execution, and result aggregation, while understanding requirements, defining test scope, assessing high‑risk operations, handling exceptions, and deciding release readiness stay with QA.

Five decision areas that must not be delegated:

Correctness of requirement and business rule interpretation

Reasonableness of test scope, priority, and risk assessment

Safety of data writes, deletions, and Redis instance selection

Handling of real‑device installation failures and abnormal results

Whether the current version meets release standards

QA Skills aim to eliminate repetitive searching, input, and operations, freeing time for genuine quality judgments.

03 Six QA Skills and Their Role in Quality Delivery

1. Test Case Design

Capability: Uses an existing test‑case generation platform to split requirements into modules, analyze test points, generate cases, and then let QA refine scenarios based on business knowledge.

Platform: Receives requirement background, functional points, constraints, and coverage goals; outputs test points and test cases.

Applicable scenarios: Post‑requirement review, modular requirement breakdown, rapid draft creation followed by QA correction.

Workflow: Requirement → module generation → test‑point generation → test‑case generation → QA refinement.

2. Smoke Quality Test

Capability: The qa-smoke-skills skill aggregates test cases, requirement docs, code paths, and change ranges to produce a checklist for smoke verification or structured code‑review reports.

Skill: Generates check points, executes smoke tests, or outputs a review report.

Applicable scenarios: Development self‑test, post‑submission quick validation of core flows, pre‑release checks of critical functions, and structured code reviews based on diffs.

Usage: Choose smoke or review mode, provide test cases, requirement docs, code paths, and change scope as input, then receive smoke results or review reports.

Precautions: Clearly define input scope, treat results as evidence rather than final judgment, and trigger at both self‑test and post‑submission stages.

3. Interface Automation

Capability: qa-interface-skills maps natural‑language requests to standardized interface‑automation scenarios, focusing on high‑frequency, well‑defined API calls.

Skill: Discovers supported scenarios, parses parameters, generates an execution plan, and runs the API after confirmation.

Applicable scenarios: Validation of login, room status, interaction status, leaderboard APIs; converting natural‑language test goals into structured requests; pre‑execution checks of scene and parameter completeness.

Usage flow: scenes → help‑scene → parse → plan → run.

Precautions: Verify scene and parameters before real execution; unsupported scenes must not skip parsing.

4. Test Data, Environment, and Result Verification

Capability: Tools for daily data inspection, cache location, and environment message observation, enabling rapid state checks, issue localization, and fault diagnosis.

Skills: qa-db-skills (natural‑language DB interaction), qa-redis-tools-skills (Redis query/modify), qa-message-printer-skill (environment message listening).

Applicable scenarios: Query test data, verify DB state, inspect Redis keys, monitor environment messages by room/user/type.

Usage: Follow a sources → parse → plan → run flow for DB operations; explicitly specify Redis instances for write/delete; define target objects and message types for message listening.

Precautions: Perform read‑only checks before writes/deletes; limit Redis multi‑instance writes to confirmed instances; restrict message‑printer to test environments only.

5. Real‑Device UI Automation

Capability: Prepares the Midscene automation environment and runs checklist YAMLs on Android/iOS devices, supporting multi‑device queues, status pages, pause, and retry.

Skills: qa-midscene-env-config (environment check/config), qa-run-android-checklist-yamls (concurrent Android checklist execution), qa-run-ios-checklist-yamls (iOS checklist execution).

Applicable scenarios: New machine setup (Node, ADB, Midscene CLI), concurrent Android device execution, iOS dual‑device runs with automatic .ios.yaml generation.

Usage: Verify environment before running; map devices to YAML lanes; ensure iPhone is connected, trusted, and Xcode/WDA/Midscene dependencies are ready.

During execution, monitor the status page for queue progress and pause or retry as needed.

6. Release Verification

Capability: Skills cover APK download/archive, Android real‑device installation and verification, and release‑mail generation, reducing manual copy‑paste of links, installations, and email templates.

Skills: qa-apk-download-skill (extract and archive APK), qa-apk-release-skill (install and verify on device), qa-release-mail-skills (draft release emails).

Applicable scenarios: Archiving APKs, installing target APKs for version and key‑check verification, generating operational or gray‑release emails.

Usage: Dry‑run download parsing, then download; ensure ADB connectivity before installation; input version, platform, and stage for email drafting.

Precautions: Download skill does not perform installation; release skill may uninstall previous app and clear data; mail skill only drafts, requiring manual review before sending.

04 Next Steps

The current article clarifies the QA Skills hierarchy, their placement in the real testing flow, and the boundaries where human judgment remains essential. Future posts will dive deeper into each skill’s implementation details, invocation process, result reporting, and concrete screenshots.

Series Overview

Test case design (refer to the linked article)

Smoke quality testing

Interface automation

Test data, environment, and result verification

Real‑device UI automation

Release verification

Subsequent articles will expand each topic with real documentation, call flows, result reports, and page captures.