QA-Exec run contract for executing authored test cases against the deployed build. Use when QA-Exec is dispatched for a task in-test to gather by-us/by-task test cases, check executable specs and instrumentation coverage, invoke the proper runner, execute UI tiers, validate visual specs, route failures, and write docs/qa-reports/<task-id>.md plus artifacts.
インストール
Codex または Claude でインストール この Prompt をコピーして Codex、Claude、または他のアシスタントに貼り付けると、Skill ページを確認してインストールできます。
QA-Exec run contract for executing authored test cases against the deployed build. Use when QA-Exec is dispatched for a task in-test to gather by-us/by-task test cases, check executable specs and instrumentation coverage, invoke the proper runner, execute UI tiers, validate visual specs, route failures, and write docs/qa-reports/<task-id>.md plus artifacts.
agents
["qa-exec"]
sdlc_phase
qa
owner
Platform Eng
status
active
QA Execution Runner
Use
Use this skill for every QA-Exec dispatch.
Inputs
Target task file and linked US/FRs
docs/test-cases/by-us/<US-ID>/ and docs/test-cases/by-task/<task-id>/
Executable specs named by each TC
docs/deploy-reports/<task-id>.md
SRS §3.4.6 Environment Configuration when the project has FE/BE runtime scope
instrumentation contract when UI selectors/testIDs are expected.
Run bidirectional instrumentation coverage for UI tasks:
every contract-declared testID appears in at least one spec;
when bundle is reachable, every contract-declared testID ships in the bundle.
Route spec gaps to QA-Author and bundle gaps to FE Dev.
Pre-run checks:
parse deploy report ## Test Environment;
verify base URLs, fixtures, env vars, build/commit/instrumentation identity;
verify env_files, env_templates, and env_validation are present; compose_config_quiet is pass; missing_required_env is none; srs_environment_contract is pass; declared_config_keys_status is all documented; and secret_values_redacted is true;
when SRS §3.4.6 declares runtime keys, verify the deploy report lists each declared key name with a non-secret status summary only; missing declared keys are a blocked environment state routed back to DevOps;
verify every Executable: file exists;
grep executable specs for TODO: instrumentation-contract; any hit is blocked.
inspect fixture reset behavior: direct DB INSERT/TRUNCATE must be followed by a synchronous reset/flush of
affected caches and process-global runtime state before the SUT is read; missing capability is blocked: test-harness-state-reset routed to BE Dev.
for direct-DB fixtures, require deploy-report test_endpoints_enabled: true and
runtime_state_reset.synchronous_probe: pass plus
runtime_state_reset.behavioral_fresh_read_probe: pass; missing or failed evidence routes to DevOps/BE Dev and
blocks the runner.
inspect runner concurrency: any global DB reset, global cache flush, or process-global reset requires per-worker
isolation or a dedicated one-worker project; otherwise mark blocked: shared-test-state.
compare navigation assertions with the signed-off URL/canonicalization contract. A path-only $ assertion that
rejects required query parameters is blocked: test-spec-contract-drift, routed to QA-Author.
if any pre-run check is blocked, halt before invoking the runner. Record the blocked reason and owning role; do
not continue to Step 7 with a known-invalid harness/spec.
Skip deprecated US test cases with reason deprecated-us; report the skip.
Select runner from solution-defaults, project ADRs, and SRS UI Introspection Profile. Load ui-test-execution and any relevant runner reference.
Invoke the runner against the deployed environment, never against production data.
For UI tasks:
Tier 1 functional always;
Tier 2 structural/token always where introspection allows;
Tier 3 visual diff only when SRS marks Visual-Critical.
Validate visual spec before UI execution:
Status: Generated;
Figma version matches confirmed task version;
handoff hash matches current handoff.
reference-render Node IDs, Figma version, dimensions, and checksums are present;
Visual Composition Assertions and Asset Assertions are non-empty and cover every handoff composition/AST row.
Report each TC as pass, fail, blocked, or skipped.
For failures, attach actual vs expected, logs/screenshots/traces/diffs, and suspected track.
Write docs/qa-reports/<task-id>.md and artifacts under docs/qa-reports/<task-id>/.
Emit plan-update.json: in-test -> done only when no fail/blocked cases remain; otherwise in-test -> failed.
UI Tiers
Tier 1 functional: required elements, copy, interactions, form validation, required states, and accessibility checks named by the SRS.
Tier 2 structural/token: layout, spacing, colors, typography, component hierarchy, and required states from docs/uiux/visual-specs/<task-id>.md.
Tier 2 asset/composition: every required logo/graphic/image/background asset is loaded from the traced implementation path and rendered with the specified fit/crop/mask/opacity/z-order; major regions and responsive variants match the Visual Composition Assertions.
Tier 3 visual fidelity: screenshot diff against approved references, only for Visual-Critical: yes.
Report Tier 2 per component/property. A component passes only when every non-skipped property in every required state passes.
No-Introspection Fallback
Use this only when SRS §3.4.3 explicitly accepts a non-introspectable UI surface, such as a Unity build where AltTester was rejected.
Confirm the SRS records the acceptance decision. If not, halt with blocked: introspection-gap.
Use the coordinate + template registry pattern:
fixed viewport/device resolution;
screenshot capture per state;
registry row per component with bounds, coordinate target, template image, source handoff node, and tolerance;
artifacts under docs/qa-reports/<task-id>/registry/.
Treat coordinate/template checks as fragile. If this is not a one-off transitional path, file a high-severity open issue recommending instrumentation.
Do not claim Tier 2 properties that cannot be observed. Mark them skipped: framework-not-introspectable.
Artifact Layout
Every run writes artifacts under docs/qa-reports/<task-id>/:
Never modify build, code, or tests to make a run pass.
Never declare done with failed or blocked cases.
Never silently skip verification.
Missing spec files, stale visual specs, build identity mismatch, and unresolved selector TODOs are blocked states.
Missing or failed deploy-report env validation, including absent SRS §3.4.6 declared-key coverage, is a blocked state routed back to DevOps. QA-Exec never guesses local .env behavior.
Flaky is failure until proven otherwise.
A serial pass after a parallel FK/cache failure does not clear the failure until the harness removes shared-state
contention or serializes the affected project by design.
Contract-incompatible test assertions are blocked QA-Author defects, not product failures; do not change the
shipped behavior to satisfy them.
Per-property UI reporting is mandatory for Tier 2.
Missing assets, placeholder substitutions, wrong frame/layout, absent backgrounds, and failed composition assertions are fe failures even when functional flows pass.
