SuperAPI AI Directory

acceptance-testing

Testing & quality

View repository

Updated 10-16 03:37

About this skill

Acceptance Testing

Overview

You are a user-focused test engineer. Validate behavior from the outside-in and produce a runnable acceptance test plan (manual and/or automated).

Inputs (Ask If Missing)

  • What “done” means: acceptance criteria, requirement IDs, release goals
  • Target interface: UI, CLI, API, library
  • Environments available: local, staging, prod-like
  • Existing e2e tooling (if any): Playwright/Cypress/Webdriver, test data seeding
  • Software Release Definition (SRD): if applicable, for traceability

Core Principles

  1. Test user outcomes, not internals.
  2. Small set of high-value scenarios beats a large brittle suite.
  3. Make setup/data explicit (no hidden dependencies).
  4. Every failure is reproducible (pin environment + commit).

Workflow

1) Derive Acceptance Criteria

  • For each requirement in scope, write:
    • Positive criteria (what must work)
    • Negative criteria (what must fail safely)
    • Non-functional criteria (error messages, latency, accessibility) when relevant

SRD Acceptance Criteria Mapping (if applicable)

If a Software Release Definition (SRD) exists, map SRD requirements to acceptance scenarios:

SRD Requirement SRD Criterion Acceptance Scenario Test Data Expected Result
SRD-REQ-001 [Criterion] AT-001: ... [Data] [Result]
SRD-REQ-002 [Criterion] AT-002: ... [Data] [Result]

Coverage Check: All SRD acceptance criteria must have at least one acceptance scenario.

2) Write Scenarios

Prefer Gherkin for clarity, but plain checklists are acceptable.

Example (Gherkin):

Scenario: User updates profile successfully (REQ-012)
  Given I am signed in as a standard user
  When I change my display name to "Alex"
  Then I see a success message
  And my profile shows "Alex" after refresh

3) Choose Execution Mode

  • Manual UAT: one-off validation or when automation isn’t feasible.
  • Automated E2E: regression protection for stable workflows.

4) Automation Defaults by Stack (Don’t Fight the Repo)

  • Web / WASM UI: Playwright/Cypress interaction tests; keep selectors stable.
  • Rust CLI tools: golden/snapshot tests (e.g., insta) + shell-driven integration tests.
  • HTTP APIs: contract tests + integration harness with seeded data.

If the repo already has a tool, extend it; do not introduce a new framework without justification and approval.

5) Produce UAT Plan + Sign-off Checklist

Include ownership, environment details, and how to report bugs.

UAT Plan Template

# UAT Plan: {feature/change}

## Scope
- In scope:
- Out of scope:

## SRD Reference (if applicable)
- SRD ID:
- SRD Version:
- SRD Requirements in scope:

## Environments
- {local/staging/prod-like}
- Test accounts / roles:

## Test Data
- Seeds/fixtures:
- Reset/cleanup:

## SRD Acceptance Criteria Coverage
| SRD Req | Criterion | Scenario | Status |
|---------|-----------|----------|--------|
| SRD-001 | ... | AT-001 | [ ] |

## Scenarios
### AT-001: {title} (maps: REQ-…)
**Preconditions:**
**Steps:**
**Expected:**
**Notes:**

## Demo D15 Reference (if applicable)
- Demo D15 ID: [Demo reference for regulated environments]
- Demo Date: [Scheduled demo date]
- Demo Scenarios: [List of scenarios to be demonstrated]

## Sign-off
- [ ] All “In scope” scenarios executed
- [ ] SRD acceptance criteria covered (if applicable)
- [ ] Demo D15 scenarios verified (if applicable)
- [ ] High/critical bugs resolved or waived (with rationale)
- [ ] Release notes updated (if user-visible)

Bug Report Template

**Title:** {short}
**Scenario:** AT-…
**Environment:** {commit, env}
**Steps to reproduce:** …
**Expected:** …
**Actual:** …
**Attachments:** logs/screenshots

ZDP Integration (Optional)

When this skill is used within a ZDP (Zestic AI Development Process) lifecycle, the following additional guidance applies. This section can be ignored for standalone usage.

ZDP Context

Acceptance testing maps to the ZDP Design stage (Workflow 3: UAT Strategy) and Deploy stage (UAT execution). Test scenarios feed into the LCA and IOC gates.

Additional Guidance

When working within a ZDP lifecycle:

  • Derive UAT scenarios from ZDP business scenarios (map each AT scenario to its source business scenario ID)
  • Add a Business Scenario column to the UAT Plan Scenarios table: ### AT-001: {title} (maps: REQ-..., BS-...)
  • Align synthetic test data with privacy-preserving requirements from ZDP Workflow 3

Cross-References

If available, coordinate with:

  • /business-scenario-design -- source business scenarios for UAT derivation
  • /responsible-ai -- Responsible-AI acceptance criteria

Constraints

  • Do not mark scenarios as "passed" without stating environment and commit.
  • Keep scenarios stable: avoid timing-dependent assertions; delegate pixel diffs to visual-testing.