J flow qa
Run full QA gate across 7 stages: lint, unit tests, NestJS E2E, Flutter integration, Playwright E2E, visual smoke check, manual checklist — layer-aware, skips stages for untouched layers. Blocks /j-flow-review if red. Usage: /j-flow-qaFrom its SKILL.md
npx -y skills add jotafierro/j-flow --skill j-flow-qaAssembled from the repository path, not quoted from the project. Check it against their README if it does not work.
One thing to look at
- 1 stars1 stars. Stars are a popularity signal and not a quality one, but at this level it is likely that nobody has read this closely except its author, and you would be relying on your own review.
SKILL.md
7.5 KB, as published. Nobody here has run it
j-flow-qa
Required reading
Override resolution: every ${CLAUDE_PLUGIN_ROOT}/… template or reference this skill reads resolves through ${CLAUDE_PLUGIN_ROOT}/skills/j-flow-shared/references/overrides.md — if a matching file exists under .specs/.overrides/, use it verbatim instead of the plugin default. For dispatching j-flow-quality, follow the agent-override dispatch rule in overrides.md §"Agent-definition overrides (dispatch)" — session confirmation and tool-scope ceiling included, never widened.
Language: prose this skill writes follows the project's **Spec language:** (and **Docs language:** where it writes under docs/ or the CHANGELOG); headings, gate vocabulary, IDs and code stay English. Canonical: ${CLAUDE_PLUGIN_ROOT}/skills/j-flow-shared/references/language-contract.md.
Before running QA, read (skip anything already in your session context from earlier this conversation):
${CLAUDE_PLUGIN_ROOT}/skills/j-flow-shared/references/gate-core.md— gate format and red/green semantics${CLAUDE_PLUGIN_ROOT}/skills/j-flow-shared/references/agent-scopes.md— j-flow-quality scope and what to test.specs/{slug}/review-guide.md— environment setup and layer file index.specs/{slug}/review/— per-layer manual testing docs (if present).specs/{slug}/tasks.json— to understand the feature scope.specs/{slug}/gate-context.md— accumulated decisions- Template:
templates/qa-report.md— output format
Gate Check
Find the active feature. Read .specs/{slug}/gate-context.md.
Require [BUILD] completed.
If missing: "Gate [BUILD] not completed. Run /j-flow-build first."
Process
Dispatch the j-flow-quality agent with:
- Full contents of
.specs/{slug}/review-guide.md - Contents of each present file in
.specs/{slug}/review/(api.md, web.md, mobile.md, admin.md, e2e.md) - Template
${CLAUDE_PLUGIN_ROOT}/skills/j-flow-shared/templates/qa-report.mdfor the output structure - Instruction: "Run all applicable QA stages in order, per the layer-scoping below. Stop on first stage failure and document exact output. Generate qa-report.md following the template: substitute test counts and status per stage, mark skipped stages as N/A with the skip reason, include exact failure output for any red stage. For the manual checklist (stage 7), use the per-layer review docs if present (see Stage 7 instructions below)."
Determine applicable stages
Read .specs/{slug}/tasks.json layers. A layer is "in scope" if it has at least one task.
- Stage 4 (Flutter integration) runs only if
mobileis in scope. - Stage 5 (Playwright E2E) runs
pnpm e2e(its own turbo task — deliberately not part ofpnpm test, and distinct fromapps/api'stest:e2eNestJS suite in Stage 3). It runs only if thee2eharness exists — i.e.apps/e2e/is present in the repo (equivalentlyhas_e2e). This is a scaffolded harness, not a build-task layer, so gate on its presence rather than ontasks.jsonlayers. When!has_web, the harness targets an externalBASE_URL; ifBASE_URLis unset, skip Stage 5 with the reason "e2e is external-target and BASE_URL is unset." - Stage 6 (Visual smoke): run the Storybook check only if
uiis in scope; run the Widgetbook check only ifmobileis in scope. If neither applies, skip Stage 6 entirely. - The
clilayer maps to Stages 1-2 only (lint + unit via vitest — already covered bypnpm lint/pnpm testacross the workspace). It triggers no NestJS E2E, Flutter, Playwright, or visual-smoke stage.
Print which stages/halves were skipped and why, alongside the stage results:
Skipped: Stage 4 (Flutter integration) — no mobile tasks in this feature.
Skipped: Stage 6 Widgetbook check — no mobile tasks in this feature.
Stages the agent runs:
Stage 1 — Lint:
pnpm lint # ESLint across all JS/TS packages
flutter analyze # Dart analyzer for mobile
Any error-level ESLint finding or Dart analyzer error is a stage failure. Warnings are logged but do not block.
Stage 2 — Unit tests:
pnpm test --passWithNoTests # or jest / vitest depending on project
flutter test
Stage 3 — NestJS E2E:
pnpm --filter @{project}/api test:e2e # runs *.e2e-spec.ts via @nestjs/testing + supertest
Requires MongoDB container. Agent checks docker compose ps first.
Stage 4 — Flutter integration:
flutter drive --target=integration_test/app_test.dart
Stage 5 — Playwright E2E:
npx playwright test
Requires the target reachable at baseURL. When a local web app exists (has_web), playwright.config.ts's webServer block boots it automatically (plus whatever it depends on — docker compose + backend). When !has_web, the harness has no webServer: ensure BASE_URL points at a running target (staging/deployed URL) before this stage — e.g. BASE_URL=https://staging.example.com npx playwright test.
Stage 6 — Visual smoke:
Storybook: pnpm storybook --ci (or equivalent) — verify it builds without errors.
Widgetbook: flutter run -d chrome --headless from widgetbook directory — verify it launches.
Stage 7 — Manual checklist:
Check whether .specs/{slug}/review/ exists.
If review/ exists (post-016 feature):
Run layer files in this order: api.md → web.md → mobile.md → admin.md → e2e.md
For each present file:
Present the layer's Checklist table to the user. Record PASS or FAIL for each row in qa-report.md.
For each row recorded PASS: flip that row's [ ] to [x] in the review file itself.
Leave FAIL rows as [ ] — failure detail is captured in qa-report.md.
review/e2e.md runs last, after all individual layer files pass.
A layer file is skipped if absent (no tasks for that layer, or infra-only).
If review/ absent (pre-016 feature):
Fall back: present each item from review-guide.md → Manual Test Steps to the user. Record PASS or FAIL.
Gate Decision
If all applicable stages pass (green):
- Write
.specs/{slug}/qa-report.mdwith green status - Append to
.specs/{slug}/gate-context.md:[QA] green {today's date} → {N} tests passing, checklist {N}/{N} - Advance the qa gate per
references/gate-core.md§"Advancing a gate" — sets the meta.md fields and recomputes the.specs/README.mdbacklog symbol. - Print the completion message, then use the Next-step dialogue from
references/gate-core.md(next command:/j-flow-review):QA gate green ✓ All applicable stages passed. Continue to next step? 1. Yes — run /j-flow-review now 2. No — stay here, I want to discuss or adjust first Enter 1 or 2:
If any stage fails (red):
- Write
.specs/{slug}/qa-report.mdwith red status and exact failure output - Append to
.specs/{slug}/gate-context.md:[QA] red {today's date} → BLOCKED: {stage name} failed — see qa-report.md - Print the completion message, then use the Next-step dialogue from
references/gate-core.md(next command:/j-flow-build --fix):QA gate red ✗ Stage failed: {stage name} See .specs/{slug}/qa-report.md for details. Continue to next step? 1. Yes — run /j-flow-build --fix now 2. No — stay here, I want to discuss or adjust first Enter 1 or 2: