manual-test-cases

Write or review manual test cases that a non-technical product owner can execute by following exact click paths. Use whenever the user says "manual test cases", "test cases for product owner", "steps

SKILL.md

GitHub

Manual Test Cases — For a Product Owner

The reader has no testing experience and no codebase knowledge. If a step can be misread, it is wrong. If a named business, button, or page doesn't exist exactly as written, the case is broken.

Scope: write cases for what the user names (module, feature); no argument = full coverage with a module table first.

Phase 1 — Map Reality

  1. Rescan the codebase (within scope) for the real module/submodule list — never trust a stale doc. Produce a module table first when coverage is the goal.
  2. Read the seeders. Build an inventory of every entity they create: names, states, relationships. This inventory is the only source of test data.
  3. Read the views for the features under test: exact button labels, exact page titles, exact locations ("Full Preview & Edit — in the card footer").
  4. When tasks/requirements.md acceptance criteria or tasks/user_journeys.md exist, they define the behaviors that must have cases — map each criterion/journey step to a case id, and flag any that can't be covered.

Phase 2 — Write Cases

Per case, this exact shape:

  • ID — module prefix + number (BIZ-07, REV-03), stable once assigned.
  • Title — states the behavior under test, unambiguous on its own.
  • Preconditions — exact setup commands (php artisan migrate:fresh --seed, which seeder class, which services must run: queue:work, schedule:work). Named data the case relies on, confirmed present in the seeder.
  • Steps — numbered clicks. Step 1 is always an entry point: a full URL ("Open http://localhost:8000/businesses") or a named case's end state. Every step names its UI element exactly as rendered: "Click Full Preview & Edit (in the card footer)". Every entity by its real seeded name.
  • Expected result — what the tester sees, concretely: badge color, banner text, row count, redirect target.
  • Result — empty PASS / FAIL: ____ line the product owner fills in during execution.

When the UI shows a failure branch (validation message, permission denial, empty state), that branch gets its own case — a PO can verify an error message as easily as a success banner.

Rejection rules — fix before delivery, never ship these

  1. Phantom data — a named entity not created by the referenced seeder. Either extend the seeder (say so explicitly as a precondition note) or rename to an existing entity.
  2. Impossible cases — steps requiring UI that is hidden in the case's state (a button hidden for suppressed records). Flag and rewrite or drop; never leave an unexecutable case.
  3. Ambiguous references — "the card", "the button", "the page" without an exact label and location.
  4. Hidden dependencies — a case depending on a prior case's side effect must name that case in preconditions ("requires BIZ-14 executed first: that business is now suppressed").
  5. Duplicates — two cases proving the same behavior; keep the clearer one.

Phase 3 — Validate the Set

Before delivery, sweep the whole file:

  • every entity name → confirmed in seeder inventory
  • every button/page label → confirmed in views
  • every case independently understandable, or its dependency explicitly named
  • consistent ID numbering, no gaps or collisions

Output Contract

  1. Cases appended/updated in tasks/manual_test_cases.md, grouped by module.
  2. A consistency note: entities validated, cases rewritten for clarity, impossible cases flagged, seeder changes required (if any).

Review Mode

When asked to review existing cases instead of writing: apply the Phase 2 rejection rules case by case, list each violation as "CASE-ID — problem — proposed rewrite", and apply fixes on approval.