SuperAPI AI Directory
Partner中文AI newsFavorites
Skillsship-check

ship-check

Testing & quality

View repository

Updated 10-29 15:01

About this skill

ship-check

A feature is done when it works in the exported desktop build and in the web build served locally, not when it works in the editor. Every check here runs the real exports, prints PASS or FAIL lines, and leaves evidence in docs/qa/<plan>/.

Read RULES.md for why each step exists. Commands are in WORKFLOW.md. The scenario list is in references/qa-scenarios.md, and the budgets in references/budgets.md.

The loop

tools/test.sh  ->  export macos  ->  --qa=<scenario> in the app  ->  export web  ->  serve  ->  browser-qa
      ^                                                                                              |
      |                     fix in code or content, never by loosening a budget                       |
      +------------------  pack check  <-  room budgets  <-  reel  <-  docs/qa/<plan>/  <-------------+
                                                                              |
                                                                     deploy.sh [--prod]
  1. Tests. tools/test.sh: headless import (fails on any script or import error), GUT, and the content lint. Logs in build/test-logs/.
  2. Desktop. tools/build/export.sh macos, then run the exported app with -- --qa=<scenario> --qa-out=<dir>. The Qa autoload plays the scenario, saves screenshots, prints [qa] PASS|FAIL lines and a final [qa] RESULT, and quits with exit code 0 or 1.
  3. Browser. tools/build/export.sh web, bun tools/build/serve.ts, then bun tools/build/browser-qa.ts <scenario>: a visible Chrome window with backgrounding off, ?qa=<scenario> in the URL, two passes (cold and warm).
  4. Pack. export.sh lists the real pack with tools/build/pck_list.py and fails if a part left out of the web build is still in it. Run python3 tools/build/pck_list.py build/web/index.pck --sizes to see what takes the space.
  5. Budgets. --qa=rooms --part=<id> tours every room: load time, frame p95, video memory with both eras and every character loaded. It fails past 400 MiB, 1 s or 16.7 ms.
  6. Reels. --write-movie with a temporary 1920×1080 override.cfg.
  7. Evidence. Screenshots, [qa] lines, measurements and reels in docs/qa/<plan>/, with a README table of checks and results.
  8. Deploy. tools/build/deploy.sh for a preview, --prod for production, then browser-qa.ts again against the deployed URL.

Quick start

tools/test.sh
tools/build/export.sh macos
build/macos/<Game>.app/Contents/MacOS/<Game> -- --qa=greybox --qa-out=$PWD/docs/qa/<plan>/greybox
tools/build/export.sh web
bun tools/build/serve.ts &                       # http://localhost:8060/
bun tools/build/browser-qa.ts greybox
QA_BASE_URL=https://<project>.vercel.app bun tools/build/browser-qa.ts greybox   # after deploy

Per-slice checklist

  1. tools/test.sh ends with == ALL PASSED.
  2. The slice's own scenario passes in the exported desktop app.
  3. The same scenario passes in Chrome on the web export, both passes.
  4. If the slice added rooms or characters: the part's room tour passes, and its numbers go in the evidence.
  5. The web pack is within its budget and holds nothing it should not.
  6. Evidence and a one-line result are in docs/qa/<plan>/README.md, and the board has an entry.
  7. After any deploy, the deployed URL passes browser-qa.ts.

Do not

  • Do not call a feature done from the editor or from GUT alone.
  • Do not loosen a budget or a QA threshold to make a check pass. Fix the content, or record a decision.
  • Do not judge frame times from a hidden or covered browser tab.
  • Do not leave override.cfg in the project after recording.
  • Do not deploy without exporting first; the export wipes build/web.
  • Do not print or upload .env or .env.local.

Back to the production map. The next part starts with story-track.