run-mazze-leczzare-blog
Skill1 starsChanged 15 days agoMITView on GitHub
Run, build, screenshot, or smoke-test mazze-leczzare-blog. Use when asked to start the dev server, verify a change works in the browser, take a screenshot of the site, or confirm a page renders correctly.
In this repository28 other files it gives its agents.
- CLAUDE.mdCLAUDE.md
- claude-md-auditSkill
- cloudflare-oneSkill
- cloudflareSkill
- component-factorySkill
- concurrency-probeSkill
- core-sampleSkill
- corroborateSkill
- decision-telemetrySkill
- disk-auditSkill
- engramSkill
- frontend-designSkill
- git-forensicsSkill
- high-signal-engineSkill
- local-swarmSkill
- model-inventorySkill
- muster-triageSkill
- precise-coding-assistantSkill
- preflightSkill
- pr-triageSkill
- read-arxiv-paperSkill
- session-journalSkill
- shipSkill
- stratum-project-initSkill
- touchstoneSkill
- transcript-handoffSkill
- web-perfSkill
- workspace-syncSkill
The file
We split this skill into the rules it gives an agent
2 rules. About 0 are common, 1 is distinctive, and 1 is very specific to this project.
≈0
Common · estimate
Many files like this have a rule of this kind.
≈1
Distinctive · estimate
This author's own choices. Worth a look.
1
Very specific
Tied to their project's names, files or tools. Probably skip.
- Common
- Distinctive
- Very specific
- Very specific: line 53
Expectations are structural, not per-page: Every route must return 200, carry a non-empty <title>, return a body of plausible size, and …
- Distinctive: line 129
Keep the asymmetry that makes it durable: the driver discovers, this file explains.
≈ “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
run-mazze-leczzare-blog
Starts the site, proves every route still serves, and hands off to the browser
for anything visual. The smoke driver (driver.cjs, next to this file) is plain
Node over HTTP — no browser binaries, no extra installs.
This file asserts as little as possible, on purpose. The previous version rotted badly: it named a project path under a macOS user that doesn't exist on this machine, claimed a framework major that had moved on twice, said there was no test suite when there are ~200 tests, told you Playwright had been removed while it sat in
devDependencies, and carried a nine-route map after the site had grown past fifty. It even documented a route as 404-in-dev that now serves 200. Every one of those was a fact written down once and never re-checked. So: derive, don't assert. The driver reads the project root, the framework version, the npm scripts and the entire route list at run time and prints them. If you are about to add a fact to this file, add a check to the driver instead.
Quick start
From anywhere inside the repo (the driver walks up to the git root itself):
node .claude/skills/run-mazze-leczzare-blog/driver.cjs --startThat starts the dev server, probes every route, prints a summary, and stops the server again. Exit code 0 means every route passed.
Other forms:
| Command | What it does |
driver.cjs | Probe a dev server you already have running (default port 4321) |
driver.cjs --port 4399 | Any port (a bare driver.cjs 4399 still works) |
driver.cjs --start | Start the server, probe, stop it |
driver.cjs --preview | npm run build, then probe the production output |
driver.cjs --base https://… | Probe a deployed site instead of localhost |
driver.cjs --expect /work/=Research | Add a substring check to one route |
driver.cjs --json | Machine-readable results on stdout |
What the driver checks, and why it's shaped that way
- Routes are discovered, never listed. It reads
dist/when a build exists — which includes dynamic routes: blog posts, projects, signal entries — and falls back tosrc/pages/otherwise. A new page is covered the moment it exists; a deleted one stops being probed. Nothing to update here, ever. - Expectations are structural, not per-page. Every route must return 200,
carry a non-empty
<title>, return a body of plausible size, and show no dev error overlay. Per-page magic strings ("expect the word Roadmap") break on every rename and train you to ignore the driver. Use--expectwhen you genuinely want one, for one run. - Environment quirks are reported, not asserted.
/rss.xmland/api/contactare probed and their status printed rather than described here. API routes are Cloudflare Pages Functions and don't run underastro dev, so they 404 locally and are real in production — the driver shows you what actually happened, which is how the old file's stale claim about/rss.xmlwould have been caught years earlier. - One real dev-vs-production difference is handled, not flagged. Pages
shipped as
public/<name>/index.html404 at/<name>/underastro dev, because the dev server doesn't resolve a directory to its index the way Cloudflare Pages does. The driver retries the explicit file and, if that serves, passes the route with a note. Without that it cries wolf on two perfectly healthy pages.
The one genuinely surprising thing: astro dev daemonizes
In the current Astro major, npm run dev returns immediately and leaves a
background dev server running. Three consequences, each verified by hitting it:
- Killing the
npmprocess does not stop the server — the port stays bound. Usenpx astro dev stop. - Astro manages one daemon per project. Asking for a second port while one
is running is a silent no-op: nothing listens on the new port and your probe
hangs until it times out. The driver checks
astro dev statusfirst and probes the running server instead of starting a doomed second one. npx astro dev logsis where the output went. The driver prints its tail when a start fails.
The driver only stops a server it started itself; one you started by hand stays up.
Visual inspection
For anything you have to look at, start the server and drive Chrome:
- Load the tools in one call —
ToolSearchwithselect:mcp__claude-in-chrome__tabs_context_mcp,mcp__claude-in-chrome__navigate,mcp__claude-in-chrome__computer,mcp__claude-in-chrome__javascript_tool,mcp__claude-in-chrome__tabs_close_mcp tabs_context_mcp→tabs_create_mcp→navigateto the local URL.- Screenshot with
computer(save_to_disk: truewhen the human should see it).
Seed the theme before load, not after. Setting
document.documentElement.dataset.theme after the page loads is overwritten by
the inline bootstrap in BaseHead and by ThemeToggle's effect — a past
session read that as a serious regression when it was a broken probe. Do this
instead, then screenshot:
localStorage.setItem('theme-preference', 'light'); location.reload();Sweep dark and light on any page you touched. Close tabs you opened.
Validation commands
Take them from package.json, not from a list here — the driver prints the
script names on every run, and the repo's own CLAUDE.md is the authority on
which gates a given change needs. As of this writing that's npm run check
before any commit, with npm test, npm run docs:check and
scripts/ops/check-docs-drift.sh alongside it.
Troubleshooting
| Symptom | Cause |
no server at … (ECONNREFUSED) after --start | Usually a daemon already running for this project on another port. npx astro dev status, then npx astro dev stop. |
| Port still bound after a run | You started that server by hand; the driver won't stop what it didn't start. npx astro dev stop. |
| A route 404s in dev but works deployed | Directory-index resolution for a public/ subfolder, or an API route that only exists in production. Re-run with --preview. |
Could not find a project root | You're outside the repo. cd into it. |
node: command not found | Node isn't on this shell's PATH. The driver prints the interpreter it ran under; any Node ≥ 18 works. |
Keeping this skill honest
Keep the asymmetry that makes it durable: the driver discovers, this file explains. Anything version-, path-, count- or page-specific belongs in code that reads it at run time and fails loudly when it changes — not in prose that goes quietly stale. A fact in this file that nothing re-checks is a future wrong answer, and this skill has already produced several.
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.

