verify-site
Skill1 starsChanged todayMITView on GitHub
Health-check the live crosbynews.com deploy with curl — every content page in BOTH languages returns 200 and is substantive, security headers present, canonical redirects are one hop, markdown negotiation works, unknown paths 404. Run after a deploy.
- Sends data out
The file
We split this skill into the rules it gives an agent
2 rules. About 0 are common, 2 are distinctive, and 0 are very specific to this project.
≈0
Common · estimate
Many files like this have a rule of this kind.
≈2
Distinctive · estimate
This author's own choices. Worth a look.
0
Very specific
Tied to their project's names, files or tools. Probably skip.
- Common
- Distinctive
- Very specific
- Distinctive: line 18
When checking that a specific change went live, sample ~8 times and require agreement before reporting either way:
- Distinctive: line 199
When everything passes, say so plainly.
≈ “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
Verify the live site
Confirm the deployed site is healthy. Use the base URL in $ARGUMENTS if one was
given, otherwise https://crosbynews.com. Run the checks below with curl, then
report a compact PASS/FAIL table. For anything that FAILs, quote the actual
status/header so it's actionable.
Deploys land in ~10–40s but take 1–2 minutes to reach every edge. A single
curl can be served by a colo still running the PREVIOUS Worker version, and a
?cb= cache-buster does NOT help — this is version propagation, not caching.
When checking that a specific change went live, sample ~8 times and require
agreement before reporting either way:
for i in $(seq 1 8); do curl -s "$BASE$PATH?cb=$RANDOM-$i" | grep -c "$MARKER"; doneMeasured 2026-08-05: 7/8 new, 1/8 old, persisting ~2 minutes past a green deploy job. Report a mixed sample as "still propagating", never as a failure.
1. Routes return 200 — both languages, every page
Sweep every content page in BOTH languages. Not a spot-check. Roughly half
the site's render branches never execute under lang="en", so an English-only
pass proves nothing about /es. This is not hypothetical: /es/hourly served a
502 in production across two deploys because features/hourly.js referenced
ES_NWS_NOTE, which only the Spanish branch reaches, and the checklist here said
"/es (Spanish spot-check)" — so /es was green while /es/hourly was down.
Take the page list from the live /sitemap.xml, never from a list written
here. It is generated from src/registry.js and holds every page in both
languages (the gauge pages under /water/… included), so it cannot fall behind
the way a hand-kept list did. Send a bot-labelled User-Agent so verification is
counted as automated traffic, not as browser visits:
UA='Mozilla/5.0 (compatible; CrosbyNewsVerifyBot/1.0; +https://crosbynews.com/)'
for u in $(curl -s -A "$UA" "$BASE/sitemap.xml" | grep -o '<loc>[^<]*' | cut -c6-); do
printf '%s %s\n' "$(curl -s -A "$UA" -H 'Accept: text/html' -o /tmp/p -w '%{http_code}' "$u")" "$u"
grep -q '<h1' /tmp/p || echo " no <h1>: $u"
done(/es is the Spanish hub, not /es/.)
Non-page routes: /robots.txt, /sitemap.xml, /llms.txt,
/api/weather, /api/health, /api/news, /api/calendar, /api/water,
/api/fishing, /api/tropics, /api/pollen, /api/burn-ban, /api/air, /api/traffic,
/alerts.xml, /news.xml, /badge.svg,
/manifest.json, /sw.js, the icon files (/favicon.ico, /favicon-32.png,
/icon-192.png, /icon-512.png, /icon-maskable-512.png,
/apple-touch-icon.png, /apple-touch-icon-precomposed.png, /og-image.png,
/notification-badge.png — all image/png except the ICO, served from
Workers static assets), /radar-image,
/.well-known/api-catalog, /openapi.json, /.well-known/security.txt,
/.well-known/mcp/server-card.json, /.well-known/agent-skills/index.json,
/.well-known/agent-skills/crosby-weather/SKILL.md.
/mcp matters more than its position suggests — it is the published MCP Registry
listing, the most externally depended-on route on the site. SKILL.md matters
because /.well-known/agent-skills/index.json publishes its SHA-256; a
regression there breaks a documented digest silently.
for p in / /weather /hourly … /es /es/weather /es/hourly … ; do
printf "%s %s\n" "$(curl -s -o /dev/null -w '%{http_code}' "$BASE$p")" "$p"
done | grep -v '^200' || echo "all 200"/api/health always returns 200 — it reports facts, not a verdict, so its
status code says only that the Worker answered. The findings are in the body:
curl -s "$BASE/api/health" | python3 -m json.toolTimestamps are Central and human-readable; each is paired with an hoursSince*
number, which is what to actually check:
curl -s "$BASE/api/health" | python3 -c "
import sys,json
d=json.load(sys.stdin)
print('cron', d['hoursSinceCronRun'], 'h ago')
for k,v in d['feeds'].items():
print(f\"{k:9} attempt={v['hoursSinceAttempt']}h ok={v['ok']} changed={v['hoursSinceChange']}h\", v.get('error',''))
"Report anything off:
ok: false— that upstream failed at its last refresh attempt. Theerrorfield says why.hoursSinceCronRunwell past 0.25 — the cron itself has stopped. Check this first: it explains every other field at once.hoursSinceAttemptlarge forweather/air/uv/water/fishing/traffic(every tick, so expect ≤0.25) means that feed stopped being attempted.airanduvare attempted withweatherand stored inside its entry, but report their ownokand content changes.calendar(~24h),tropics(~1h in season / ~24h off-season),pollen(weekday mornings only), andburnban(~4h) are throttled, so lagging is normal for them;newsisnullby design (written out-of-band).hoursSinceChange>>hoursSinceAttempt— the important one. Refreshes are succeeding while the content sits frozen, which is exactly how/pollenserved a three-day-old count with every other signal green. There is no single threshold; judge it against how often that feed's data genuinely moves — hours for weather/water/traffic, a day for pollen and burn-ban status, longer for calendar and a quiet tropics basin.newslegitimately goes quiet in a small town.
Also assert each page is substantive, not just 200. A page that renders but lost its data reads as healthy on status alone:
curl -s "$BASE$p" | wc -c # content pages are >2KB; a bare error page is far smaller2. Security + negotiation headers on /
curl -sI "$BASE/" and confirm each header is present:
strict-transport-securityx-frame-optionscontent-security-policyx-content-type-options: nosniffreferrer-policypermissions-policyvary: Acceptlink:(advertises the markdown alternate, sitemap, api-catalog, openapi)
3. Canonicalization — ONE hop each
Without following redirects (curl -sI), each variant must 301 straight to the
apex https://crosbynews.com/... (query string preserved) in a single hop. Read
the location: header:
http://crosbynews.com/→https://crosbynews.com/https://www.crosbynews.com/→https://crosbynews.com/http://www.crosbynews.com/→https://crosbynews.com/(still one hop)
https://crosbynews.com/ itself must NOT redirect (expect 200).
http://www false-fails from a session. Plain-HTTP requests to
www.crosbynews.com intermittently return a 503 with no cf-ray/server:
cloudflare header — that's the sandbox's outbound proxy failing to relay, not
the site (measured 2026-07-28: ~6 of 8 attempts, while http:// apex was 12/12
and https://www 6/6 clean, and a GET that does get through returns
Cloudflare's real 301 page). Don't report it as a redirect regression. Re-run,
confirm the same path over https://www, or check from outside the sandbox.
Parsing gotcha in this environment: curl -sI here returns an extra
HTTP/1.1 200 Connection Established CONNECT line before the real response
(the session's outbound HTTPS proxy inserts it), so naively taking the first
HTTP/... line reads that CONNECT line's 200, not the actual status — a
real 301 can misread as 200. Skip it explicitly:
grep "^HTTP" | grep -v "Connection Established" | head -1. Also strip the
header value's trailing \r before an exact string comparison (| tr -d
'\r') — curl includes it raw, so [ "$location" = "..." ] silently fails
even when the printed values look identical. Simplest fix: prefer
curl -s -o /dev/null -w "%{http_code}" (as in check 1) over parsing -sI
output when you only need the status code — it isn't affected by the CONNECT
line at all.
4. Markdown content-negotiation
/ should return markdown (not HTML) when asked two ways:
curl -s "$BASE/?format=md" | headcurl -s -H "Accept: text/markdown" "$BASE/" | head
5. Unknown path 404s
curl -s -o /dev/null -w "%{http_code}\n" "$BASE/this-path-does-not-exist" # expect 4046. Content pages are GET/HEAD only — and the two /mcp paths are not
Content pages answer any other method with 405 and Allow: GET, HEAD. The
exemptions matter more than the rule: POST /mcp is the JSON-RPC protocol and
must keep working, and POST /es/mcp must keep answering 404, which
/developers documents in both languages. A guard that swallowed either would
break the site's most externally depended-on route while every page still looked
fine.
curl -s -o /dev/null -w "%{http_code} PUT /weather\n" -X PUT "$BASE/weather" # 405
curl -s -o /dev/null -w "%{http_code} DELETE /es\n" -X DELETE "$BASE/es" # 405
curl -s -o /dev/null -w "%{http_code} POST /unknown\n" -X POST "$BASE/nope" # 404, NOT 405
curl -sI -X PUT "$BASE/weather" | tr -d '\r' | grep -i '^allow:' # allow: GET, HEAD
# the exemptions
curl -s -X POST "$BASE/mcp" -H 'content-type: application/json' \
-d '{"jsonrpc":"2.0","id":1,"method":"tools/list"}' | head -c 60 # a tools list
curl -s -o /dev/null -w "%{http_code} POST /es/mcp\n" -X POST "$BASE/es/mcp" # 404A 405 is still a normal response here: it carries the security headers like everything else, so check one if section 2 is in doubt.
When everything passes, say so plainly. Otherwise list only the failures with the observed value next to the expected one.
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.

