QA Acceptance Agent Skill

QA Acceptance

Write acceptance criteria that a reviewer can run against the running app and decide pass or fail without asking the author. The criteria are the contract — automated tests cover correctness, QA covers feature-level behavior.

When to use

• A feature change is heading to QA and needs a written validation plan.
• A reviewer is asked to verify a PR that touches user-visible behavior.
• An incident postmortem requires a regression check before reopen-prevention.
• A release candidate needs a pre-cut smoke pass.

When not to use

• The change is unit-test-only (utility refactor, internal naming). Acceptance criteria are unnecessary churn.
• You are asked to write tests against API contracts. Use contract testing, not feature QA.

Acceptance criteria format

Each criterion is a single, independently-verifiable statement:

- **Given** <starting state>, **when** <action>, **then** <observable outcome>.

Example:

- **Given** a CSV export with 0 rows, **when** the user clicks Export, **then** the file downloads with only the header row and the UI shows "Exported 0 rows".

Avoid criteria that combine multiple whens or thens. Split them.

What every plan must cover

• Golden path. The most common successful flow, end to end.
• Empty and minimum states. Zero items, one item, missing optional inputs.
• Boundary inputs. Max length strings, max numeric values, unicode, RTL text where applicable.
• Error states. Network failure, permission denied, validation failures, conflict (409), not found (404).
• Concurrency and ordering. Two users acting at once, race against background jobs, refresh during mutation.
• Performance envelope. The largest realistic input the change must handle without UI hangs or timeouts.
• Backward compatibility. Existing data, existing URLs, persisted user preferences continue to work.
• Telemetry and audit. Events, logs, or activity entries the change is supposed to emit.

If a section is genuinely not applicable, write "N/A: <why>" — do not silently omit.

Evidence

Each criterion needs evidence on the verification pass:

• Screenshot or short clip for UI behavior.
• Copied console / network output for API behavior.
• Log snippet or activity row for telemetry.
• Timing measurement for performance criteria.

"Looks good to me" without evidence is not a pass.

Quarantine and follow-up

• A failing criterion blocks acceptance unless explicitly waived by the owner with a tracked follow-up issue.
• "Known issue" without a linked follow-up is not a waiver.
• If you add a new criterion mid-pass, restart the pass — partial coverage hides regressions.

Handoff back to the author

Return the validation plan with three sections:

• Pass. Criteria that passed, with one-line evidence summaries.
• Fail. Criteria that failed, with the exact reproduction.
• Blocked. Criteria you could not run, with why.

The author owns turning failures into either fixes or accepted deferrals.

Anti-patterns

• Acceptance phrased as test plan ("write a Cypress test for X"). Acceptance is what is true after the change ships; tests are how you check.
• Criteria that depend on inspecting implementation details (selectors, query plans). Stay observable.
• Long checklists with no priority. Mark must-pass criteria distinctly from nice-to-have.
• Validation reports that say "passed" with no evidence. Reviewers cannot audit those.