agentsclimarketplace

J flow qa

Skill jotafierro/j-flow/skills/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

Install
npx -y skills add jotafierro/j-flow --skill j-flow-qa

Assembled 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):

  1. ${CLAUDE_PLUGIN_ROOT}/skills/j-flow-shared/references/gate-core.md — gate format and red/green semantics
  2. ${CLAUDE_PLUGIN_ROOT}/skills/j-flow-shared/references/agent-scopes.md — j-flow-quality scope and what to test
  3. .specs/{slug}/review-guide.md — environment setup and layer file index
  4. .specs/{slug}/review/ — per-layer manual testing docs (if present)
  5. .specs/{slug}/tasks.json — to understand the feature scope
  6. .specs/{slug}/gate-context.md — accumulated decisions
  7. 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.md for 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 mobile is in scope.
  • Stage 5 (Playwright E2E) runs pnpm e2e (its own turbo task — deliberately not part of pnpm test, and distinct from apps/api's test:e2e NestJS suite in Stage 3). It runs only if the e2e harness exists — i.e. apps/e2e/ is present in the repo (equivalently has_e2e). This is a scaffolded harness, not a build-task layer, so gate on its presence rather than on tasks.json layers. When !has_web, the harness targets an external BASE_URL; if BASE_URL is 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 ui is in scope; run the Widgetbook check only if mobile is in scope. If neither applies, skip Stage 6 entirely.
  • The cli layer maps to Stages 1-2 only (lint + unit via vitest — already covered by pnpm lint / pnpm test across 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):

  1. Write .specs/{slug}/qa-report.md with green status
  2. Append to .specs/{slug}/gate-context.md:
    [QA] green {today's date}
      → {N} tests passing, checklist {N}/{N}
    
  3. Advance the qa gate per references/gate-core.md §"Advancing a gate" — sets the meta.md fields and recomputes the .specs/README.md backlog symbol.
  4. 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):

  1. Write .specs/{slug}/qa-report.md with red status and exact failure output
  2. Append to .specs/{slug}/gate-context.md:
    [QA] red {today's date}
      → BLOCKED: {stage name} failed — see qa-report.md
    
  3. 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:
    

Keep looking

Skills are one crate of 608,475. Ordering is by how many stacks a row turns up in, so the top of any crate is what has actually been picked rather than what has the most stars.