Skip to content

Migrated from Notion GrokBot role — https://app.notion.com/p/3d31f8eaa3ad815584caf54537cf0f12

GrokBot role (QA Engineer)

Callout Who should read: anyone using the QA Engineer GrokBot, or wiring oracle/Playwright gates. Callout As of 2026-09-24 IST — stack now includes Karate+k6 manual workflow and Project 10 board. Tips: API develop 94f64f5; browser main 4a53eab. Coverage audit: TEST coverage recommendations (2026-09-24). Project 10. How to use: GrokBot invents + gathers evidence; CI owns pass/fail. Chat is never the gate. Next: Test approach · Scenario matrix · TEST coverage recommendations (2026-09-24). Audience: anyone running BookingZone / VenuPlus quality with Anil’s QA Engineer GrokBot. Last updated: 2026-09-24 IST. Related: bz-platform-context#107, oracle in reservation-api-server/tests/sys005_oracle/, Karate/k6 under tests/karate-api-tests/ + tests/k6-load-tests/ + workflow sys005-karate-k6-manual.yml, Playwright in reservation-browser-tests, board Project 10.

One-line rule

GrokBot invents scenarios and gathers evidence; pytest / oracle / Playwright / Actions own the gates. Do not treat chat pass/fail as CI.

What GrokBot does

Does Does not
Invent / refine class-bound cases (Gherkin, matrix rows) from live config Replace the offline oracle fixtures
Drive DEV UI/API read-mostly smokes and capture screenshots Be the merge gate for product PRs
Write Joe Colantonio–style reports: expected vs actual, ship / no-ship Charge production Datacap unless Anil explicitly approves
Compare live behavior to oracle helpers (clamp_guests, expected_end, hold_window, deposit/promo) Add new tests for a new Outlet of an existing class_id (class-bound = zero)
Keep local markdown/JSON under bz-qa/ as source of truth; refresh OneDrive Word/Excel share packs Hand-edit Excel as the system of record
Propose board hygiene / evidence for Epic H packs already on develop Invent SKUs or raise k6 above 1 VU without approval
## Quality stack (who owns what)
1. Oracle (CI, cheap) — reservation-api-server → tests/sys005_oracle/ + workflow sys005-oracle.yml. Booking-math fixtures; ≥30 cases; no browser/DEV API. Merged via #925 to develop as CI Quality Gate.
2. Unit (ci-quality.yml) — ruff + pytest tests/unit/ fail-closed on PR. Does not yet include tests/security/ (S-IDOR #163 orphaned).
3. Karate + k6 (manual) — #974 packs on develop; workflow sys005-karate-k6-manual.yml (docs-only / k6-inspect / karate-dryrun). Concurrency/exhaust @manual-only. Epic H #166–#169 still Backlog pending operator evidence — not new scenarios.
4. Playwright (Actions) — aerospace-apps/reservation-browser-tests. Tip 4a53eab (2026-09-08). Nightly = suite=admin + business_admin only; often red (outlet select). Buyout / pay-later / slot-full / PWA = dispatch or gated.
5. GrokBot evidence — ship/no-ship reports with screenshots + expected vs actual (not pytest pass counts). After deploys or when Anil asks.
## Invent vs gate
- Invent = propose scenarios, fill gaps, draft Gherkin, point at missing coverage. Output is a proposal until it lands in oracle fixtures, Playwright specs, Karate features, or the board.
- Gate = green oracle + unit on PR, green browser Actions when relevant, plus Anil’s read of a ship/no-ship note when the change touches live booking pay/create.
## Standing constraints
- Org: aerospace-apps. Master spec: bz-platform-context#107. Board: Project 10.
- Ask before GitHub writes (except when Anil already greenlit a track).
- No update_booking on Confirmed locks; desk path is admin_booking.patch_booking.
- New Outlet of an existing config class = zero new tests.
- Prefer class goldens (Midway / Super Pass / ULTIMATE); buyout is documented non-golden.
- No live Datacap / destructive flags without Anil approval.
- Prefer Colantonio-style evidence reports over pytest pass counts.
## Share locations
- OneDrive (Hawktive): Documents/Projects/BZ — living Word/Excel/markdown copies.
- Machine source of truth: local bz-qa/phase0/knowledge-base/ + repo oracle under reservation-api-server/tests/sys005_oracle/.
- Notion hub: Quality (SYS-005).
## Sample report shape
## <change or smoke title>
Expected: …
Actual: …
Evidence: <screenshot paths / booking ids>
Conclusion: SHIP | NO-SHIP
Notes: <follow-ups only>