post-deploy-smoke
Skill0 starsChanged 22 days agoMITView on GitHub
Smoke-test a deploy on the LIVE URL in real browsers (Chromium 1440 and WebKit/Safari 390) and fail on what a status code hides - 4xx/5xx on any asset, JS errors, images that loaded but are invisible or never decoded, reveal-on-scroll content still hidden, buttons cut off at the phone edge, sideways scroll. Use right after every deploy or release of a website or web app, before saying "deployed", "live", "done" or sending anyone a link, and when the user says "smoke test", "check the deploy", "is the site OK", "verify production", "post-deploy check" or "did it break anything".
- Installs packages
In this repository4 other files it gives its agents.
- multi-session-gitSkill
- permission-unblockSkill
- prove-its-liveSkill
- secret-guardSkill
The file
We split this skill into the rules it gives an agent
8 rules. About 0 are common, 6 are distinctive, and 2 are very specific to this project.
≈0
Common · estimate
Many files like this have a rule of this kind.
≈6
Distinctive · estimate
This author's own choices. Worth a look.
2
Very specific
Tied to their project's names, files or tools. Probably skip.
- Common
- Distinctive
- Very specific
- Distinctive: line 19
After every production deploy, on every URL whose code, content, CSS or assets changed, for every site you touched.
- Distinctive: line 20
Before deploying, on the preview or local build, so the problem never ships.
- Distinctive: line 21
Before you write "deployed", "live", "fixed in prod", or send a link to anyone.
- Very specific: line 63
Wait until the deploy is actually serving your commit (see the prove-its-live skill), then run the script on the live URLs.
- Distinctive: line 69
Never report "deployed" or "working" off a build log, a 200, or a local run.
- Distinctive: line 71
If the script could not run (exit 3, network down), say "not verified", not "looks fine".
- Very specific: line 72
Do not silence a finding with --ignore unless the URL is genuinely third-party and outside your control.
- Distinctive: line 73
When a human finds a bug this missed, add a check for it and prove it catches the old version.
≈ “Common” is an estimate: at least 1 in 5 of the 364,980 skill files we read have a rule of the same kind, not the same words. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.
Files
Post-deploy smoke test
A deploy is not finished when the build is green or curl returns 200. It is finished when every page you
touched loads on the live address in a clean browser with no JS errors and everything that should be
visible is visible. 200 from the server is not "the image is visible."
Real failures this catches: a new HTML shipped next to a cached old app.js, the script threw, and every
image stayed at opacity: 0 while the server answered 200 for all of them. A "fixed" button that was fine on
desktop and half off-screen on every phone.
When
- After every production deploy, on every URL whose code, content, CSS or assets changed, for every site you touched.
- Before deploying, on the preview or local build, so the problem never ships.
- Before you write "deployed", "live", "fixed in prod", or send a link to anyone.
Run it
Needs Node 18+ and Playwright (npm i -D playwright && npx playwright install chromium webkit). If Playwright
is missing the script exits 3 and says so; nothing was checked, so do not report a pass.
node ${CLAUDE_SKILL_DIR}/smoke.mjs https://example.com/ https://example.com/pricing${CLAUDE_SKILL_DIR} is this skill's folder (the folder holding this SKILL.md). Options:
| Option | Use |
--engines chromium,webkit | default both: Chromium 1440x900 and WebKit 390x844 as an iPhone |
--ignore <regex> | skip failing third-party URLs you do not own (analytics are skipped by default) |
--reveal <selector> | your own reveal-on-scroll class if it is not .reveal, [data-aos], .fade-in... |
--wait <ms> | settle time after load and after scrolling (default 1500) |
--json | machine-readable output |
Exit 0 = every page clean in every engine. Exit 1 = something failed. Exit 2 = bad arguments. Exit 3 = Playwright or a browser missing.
What it checks (each page, each engine, in a fresh context)
| Kind | Meaning |
http | any response >= 400: the page, an image, a script, a font, an API call |
request-failed | DNS, TLS, refused or blocked requests |
js-error / console-error | uncaught exceptions and console.error |
broken-image | <img> finished loading with naturalWidth 0 (404, wrong type, corrupt, HTML served as an image) |
invisible-media | image or video that is rendered but has opacity 0 or visibility: hidden on it or a parent, after scrolling the whole page (carousel slides waiting their turn are allowed when another slide is visible) |
unrevealed | reveal-on-scroll blocks still hidden after the whole page was scrolled |
hidden-but-shown | element with the hidden attribute that CSS draws anyway |
dead-video | <video> with no data |
cut-off | link, button or field crossing the screen edge (not inside a scroll container) |
sideways-scroll | the page is wider than the viewport |
Procedure
- List the URLs that changed (from the diff: routes, templates, shared CSS/JS means "every page type"). Include pages nothing links to (404, legal, a product page).
- Wait until the deploy is actually serving your commit (see the
prove-its-liveskill), then run the script on the live URLs. - For each
FAIL: open the page yourself, find the cause in the code, fix it, deploy, and run the script again on the live URL. - Report per site: what was checked (pages, engines), what failed and was fixed, what remains and why, and the full live URL.
Rules
- Never report "deployed" or "working" off a build log, a 200, or a local run. Only off this script passing on the live URL in this session.
- A pass on
localhostor a preview URL is not a pass on production: caches, CDNs, env vars and redirects differ. - If the script could not run (exit 3, network down), say "not verified", not "looks fine".
- Do not silence a finding with
--ignoreunless the URL is genuinely third-party and outside your control; say which ones you ignored. - When a human finds a bug this missed, add a check for it and prove it catches the old version.
Does it work?
Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.
No reviews yet
People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.
- It worked
- Partly
- It didn't
Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.
Your trail
Files you open will collect here, so you can find them again.
Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.
Now do this for your own.
Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.

