Terminal
Terminal/Skill
Checks a web page on a schedule with Tavily and reports only when its content meaningfully changes, such as a pricing page or changelog.
By Tavily
Works with
npx skills add tavily-ai/use-case-skills --skill watch-for-changesCheck something on the web on a recurring cadence, compare it to what you saw last time, and only report back when there's an actual change.
If tvly is not found on PATH, see https://github.com/tavily-ai/tavily-cli for installation instructions.
Tavily's API is stateless — it has no built-in scheduling, alerting, or "watch this" endpoint. This skill combines:
tvly extract or tvly search) for the actual "check the thing" stepschedule skill (cron-based, unattended) or /loop (self-paced, session-scoped)schedule trigger)Be upfront with the user about this — "monitor" here means "something re-runs this check for you on a cadence," not "Tavily is watching this in the background on its own." Ask the user which mechanism is available/preferred if it isn't obvious.
tavily-news-digest idea in this repo's suggestions doc. This skill is for detecting change, not aggregating new content.schedule skill or /loop, another agent's own scheduler, or plain cron/CI if running standalone).1. Take the baseline snapshot:
python3 << 'PYEOF'
import json, subprocess, hashlib, os
url = "https://example.com/pricing"
raw = subprocess.check_output(['tvly', 'extract', url, '--json'], stderr=subprocess.DEVNULL)
data = json.loads(raw)
content = data['results'][0]['raw_content']
os.makedirs('.tavily', exist_ok=True)
snapshot = {
"url": url,
"content": content,
"hash": hashlib.sha256(content.encode()).hexdigest(),
}
with open('.tavily/monitor_pricing.json', 'w') as f:
json.dump(snapshot, f)
print("Baseline saved.")
PYEOF
2. Set up the recurring check — point whatever recurring-task mechanism is available at the diff script below: the schedule skill or /loop <interval> in Claude Code, an equivalent scheduling primitive in another agent, or a plain cron entry / scheduled CI job if running standalone.
3. Each scheduled run — diff against the saved snapshot:
python3 << 'PYEOF'
import json, subprocess, hashlib
url = "https://example.com/pricing"
path = ".tavily/monitor_pricing.json"
with open(path) as f:
prev = json.load(f)
raw = subprocess.check_output(['tvly', 'extract', url, '--json'], stderr=subprocess.DEVNULL)
data = json.loads(raw)
content = data['results'][0]['raw_content']
new_hash = hashlib.sha256(content.encode()).hexdigest()
if new_hash != prev['hash']:
print(f"CHANGED: {url}")
# Optionally diff prev['content'] vs content and summarize the change in plain English
# (e.g. via a quick tavily-research call or a direct text diff) before alerting the user.
else:
print(f"No change: {url}")
with open(path, 'w') as f:
json.dump({"url": url, "content": content, "hash": new_hash}, f)
PYEOF
| Consideration | Guidance |
|---|---|
| Cadence | Match check frequency to how often the thing actually changes — a pricing page doesn't need hourly checks; a live filing tracker might |
| Snapshot storage | Keep one JSON file per monitored target under .tavily/, named for what it watches |
| Noisy pages | If a page has content that changes trivially every load (timestamps, ad slots), hash a specific section's content, not the whole page — extract with --query/--chunks-per-source to narrow to the relevant section first |
| What changed | For a human-readable summary of what changed (not just that it changed), diff the old and new text and optionally run it through tvly research for a plain-English summary |