Agent Skills: Test Scenarios

Test Scenarios

Generate complete scenario coverage from a feature spec before tests are written. Closes the gap between "we built it" and "it survives production."

When to use this skill

• After PRD approval but before engineering implementation
• During sprint planning to size testing effort
• During code review to verify test coverage
• During QA planning to scope test pass
• During bug-bash prep
• After a production incident to validate scenario gaps

The 7 scenario categories

For every feature, generate scenarios across:

1. Happy paths — primary user goals achieved cleanly
2. Edge cases — boundary conditions, unusual inputs
3. Error handling — what users see when things go wrong
4. Empty states — first-use, no data, after-delete
5. Concurrent operations — race conditions, optimistic locking
6. Accessibility — keyboard nav, screen reader, contrast, motion
7. Security + privacy — auth, permissions, PII, injection

Plus when applicable:

• Performance scenarios (load, latency, throughput)
• Localization (RTL, long-string, currency, date format)
• Cross-platform (browsers, devices, OS versions)

Clarify First

Before generating scenarios, confirm these inputs. If any is unknown or vague, ASK — do not assume:

• [ ] Feature type — form, browse/list, real-time/collab, file upload, payment, or bulk operation (drives the scenario count per category in the coverage rubric)
• [ ] Risk / sensitivity profile — auth, payment, or PII involved (scales up security and concurrency coverage in the risk-weighted selection)
• [ ] The spec's inputs, outputs, and side effects — what users provide, expect, and what changes in the system (drives edge cases in Step 3 and error handling in Step 4)

Stop rule: ask only the 2-3 that most change the output. If the user says "just draft it," proceed and list your assumptions at the top of the artifact.

Workflow

Step 1 — Spec read

Read the PRD / user story / acceptance criteria. Identify:

• User goals (what they want to do)
• Inputs (what they provide)
• Outputs (what they expect)
• Side effects (what changes in the system)

Step 2 — Happy path

For each user goal, write the primary flow:

• Preconditions
• Steps
• Expected result

Aim for 1-3 happy paths per feature.

Step 3 — Edge cases

For each input, ask:

• What's the empty value?
• What's the minimum?
• What's the maximum?
• What's just below min / just above max?
• What's the wrong type?
• What's the weird-but-valid (very long, special chars, Unicode, emoji)?

Step 4 — Error handling

For each failure mode:

• Network failure
• Server error (5xx)
• Validation error (4xx)
• Timeout
• Concurrent modification
• Auth expired / lost

For each: what does the user see? Recover-from / try-again UX?

Step 5 — Empty / first-use / after-action states

• First use (no data)
• After delete (last item)
• After error (partial state)
• After timeout
• After cancel

Step 6 — Concurrent / race scenarios

Two users simultaneously:

• Editing same record
• Triggering same action
• Submitting same form
• Reaching capacity limit

Step 7 — Accessibility

• Tab through with keyboard
• Use with screen reader
• High contrast / inverted colors
• Reduced motion preference
• Large text scaling
• Voice control

Step 8 — Security + privacy

• Logged out user accesses
• Unauthorized user accesses
• Permission downgrade mid-action
• PII handling
• Injection attempts (XSS, SQLi)
• Rate limiting

Step 9 — Run test_scenario_generator.py

Audit a candidate scenario list for category coverage; flag gaps.

python3 project-management/execution/test-scenarios/scripts/test_scenario_generator.py \
  --input feature_spec.json --format markdown

Decision frameworks

How many scenarios per category?

| Feature type | Happy | Edge | Error | Empty | Concur | A11y | Security | |--------------|-------|------|-------|-------|--------|------|----------| | Form / submission | 1-3 | 4-8 | 4-6 | 2 | 1-2 | 4 | 3-5 | | Browse / list | 2-3 | 3-5 | 2-3 | 2-3 | 1 | 3 | 2 | | Real-time / collab | 3 | 4-6 | 4-6 | 2 | 4-6 (essential) | 3 | 3 | | File upload | 2 | 6-10 (sizes/types) | 4-6 | 1 | 1-2 | 2 | 5+ (file abuse) | | Payment / financial | 3 | 6-10 | 8+ (critical) | 2 | 4-6 (idempotency!) | 3 | 8+ | | Bulk operation | 2 | 4-6 (sizes) | 4-6 | 1 | 2-4 (partial fail) | 2 | 3 |

Adjust for risk profile of the specific feature.

Risk-weighted scenario selection

Not all scenarios need full QA coverage. Apply:

• Happy path: always test
• Edge cases: test those that matter (likelihood × severity)
• Error handling: test all paths that user can recover from
• Empty state: always test (first impression!)
• Concurrent: test for data-integrity-critical features
• Accessibility: baseline coverage on all UI; full coverage on user-facing
• Security: scale with sensitivity (full coverage on auth/payment)

Manual vs automated

| Scenario type | Default | |----------------|---------| | Happy path | Automated (E2E or integration) | | Edge cases (input validation) | Unit tests | | Error handling | Mix (mocked errors in unit; real in integration) | | Empty state | Visual regression + manual | | Concurrent | Hard — usually manual + targeted integration | | Accessibility | Automated (axe-core) + manual screen-reader | | Security | SAST + DAST + manual review for critical paths | | Performance | Automated load tests | | Localization | Pseudo-localization + manual spot-check |

Common engagements

"Generate test scenarios for this new feature"

1. Read the PRD / spec.
2. Apply the 7 categories.
3. Generate scenarios; estimate count per category.
4. Run validator; address gaps.
5. Hand to QA for implementation.

"Audit our test plan for completeness"

1. Categorize existing scenarios.
2. Identify under-covered categories.
3. Score by risk × likelihood.
4. Recommend additions.

"Post-incident scenario gap analysis"

1. Pull the incident scenario.
2. Map: which category was missed?
3. Add scenario; verify regression coverage.
4. Update default-coverage rubric for that feature type.

Anti-patterns to avoid

• Only happy paths. Production fails on edges; you skipped them.
• No empty state. First-use is broken.
• "QA will figure it out." No, QA tests what's specified.
• Generic acceptance criteria. "Should work" is not testable.
• No accessibility scenarios. Excludes users; fails compliance.
• No security scenarios. Vulnerable paths ship.
• No performance scenarios for performance-sensitive features. Surprise at scale.
• Skipping concurrent for collaborative features. Race conditions ship.

References

• references/scenario-categories.md — deep on the 7+ categories with examples
• references/coverage-anti-patterns.md — common gaps + fixes

Related skills

• project-management/execution/create-prd — upstream spec
• project-management/execution/wwas — acceptance criteria
• engineering/senior-qa — implementation
• engineering/code-reviewer — review coverage
• product-team/spec-to-repo — translating spec to tickets