Earn 14 free days when your bug report or suggestion is accepted — how it works

Claude Code multi-app test pack

This page is the public version ofdocs/CLAUDE_CODE_TEST_PACK.mdin the repo. The pack gives you 5 starter projects and 5 prompts to paste into Claude Code — one per stack — so you can see CodeLoop's verify → diagnose → fix → gate-check loop drive every common codebase shape.

Bootstrap the fixtures

bash scripts/claude-code-app-pack/clone-fixtures.sh

This shallow-clones each starter into claude-code-fixtures/<app-id>/ and writes a pre-seeded .codeloop/config.json + CODELOOP_FIXTURE_README.md. Use --target /custom/dir for a different output location or --only nextjs-playwright for a single fixture.

The 5-app matrix

  • nextjs-playwright — Next.js + Playwright web app. Add an /about page with a heading and a Playwright test that asserts the heading. Expected: visual_review, interaction_replay, gate_check ≥ 94 %.
  • flutter-desktop — Flutter desktop sample. Add a Theme menu (light / dark) with persistence + widget test. Expected: 2 screenshots (light + dark), video of toggle.
  • macos-swift— Native macOS Swift app. Add a Settings pane with a “Reset cart” button + unit test. Expected: settings + alert screenshots, video of full reset flow. macOS-only.
  • express-api — Node.js Express REST API. Add a GET /v1/health endpoint + supertest case. API row — no screenshots / video required; the pre-seeded config sets ui_review_required: false.
  • django-react — Hybrid backend + frontend. Add a /api/items endpoint backed by a Django model and a React <ItemList> component + tests on both sides.

Common workflow

  1. Open the fixture folder in Claude Code.
  2. Paste the prompt from the test-pack doc in the chat panel.
  3. Watch the assistant call codeloop_init_project → codeloop_verify.
  4. Capture screenshots and video as the prompt instructs.
  5. Confirm codeloop_gate_check returns ready_for_review with confidence ≥ 94%.
  6. Save evidence to tier-b-evidence/claude-code/<app-id>-<timestamp>/.

Evidence appears in codeloop doctor

After any capture, npx codeloop doctor enumerates the Claude Code app IDs that have folders under tier-b-evidence/claude-code/ on this host, and lists any missing apps with one-liner remediation. Honest gating: missing rows are surfaced but never PR-block, because the captures are user-executed.

For the per-app prompt text, expected MCP tool-call sequence, and evidence checklist, see docs/CLAUDE_CODE_TEST_PACK.md in the repo. The matching rows in the master E2E checklist are 25.13–25.17.