anakin-inc/agent-skills/anakin-monitoring

Anakin: website monitoring

Scheduled checks that record a change when watched content differs from the previous run, optionally alerting a webhook or email.

Use this instead of re-scraping on a loop. If the user says "tell me when", "watch for", "alert me if", or "track over time", that is a monitor — not a repeated scrape.

Creating a monitor

monitor_create requires url and intervalMinutes (minimum 15).

Scope — what gets watched

• "page" (default) — one URL.
• "site" — crawls the site each run and tracks pages added, removed, changed.

Scope it with maxPages, maxDepth, includePatterns, excludePatterns.

• "wire" — runs a Wire action each check and diffs its JSON. Pass

wireActionId (or wireCatalogSlug), wireParams, wireCredentialId, and wireWatchPaths to watch specific fields.

Watch mode — this is the cost lever

• "full_page" — 2 credits/check, compares the whole page.
• "specific_data" — 3 credits/check, extracts only the fields in

outputSchema with AI.

specific_data costs more per check but is usually the right choice for price, stock, or status tracking: full_page fires on any incidental edit, so you pay for checks and wade through noise. Define an outputSchema with just the fields that matter.

aiMode adds +1 credit/check and filters trivial noise, summarizing real changes. Worth it on pages that churn. Pair it with aiGoal, a plain-language rule for which changes count, e.g. "only when the price drops or it goes out of stock". Without one, the filter guesses what matters.

In full_page mode, watchFormat sets what is compared: "markdown" (default), "html", or "cleaned_html". Keep the default unless the change you care about lives in markup that markdown drops, such as attributes or hidden elements.

Cost arithmetic before you create

Credits/day = (1440 / intervalMinutes) x per-check cost. A 15-minute specific_data monitor with aiMode is 96 x 4 = 384 credits/day. Compute this and tell the user before creating — a short interval on a high-cost mode is an easy way to burn a plan.

Active-monitor caps: Free 5, Pro 20, Scale 100.

Other options

alertWebhookUrl, alertEmails, useBrowser, country, sessionId (for login-protected pages — see anakin-browser), expiresAt, isActive.

Managing monitors

• monitor_list — all monitors, or pass id for one monitor's full config and

status (next/last check, active state, per-check cost, alerts). Call this first to get an id.

• monitor_changes — the detected changes for a monitor id, each with a

diff/summary (plus the AI summary when aiMode is on).

• monitor_control — "pause", "resume", "run_now", "delete".
• resume may hit the plan's active-monitor cap.
• run_now is billed like a normal check.
• `delete` permanently removes the monitor and its entire history. Confirm

with the user first; there is no undo.

Working pattern

1. scrape the URL once to confirm the data is actually there and stable.
2. Decide watchMode from what the user cares about — define outputSchema for

specific_data.

1. Pick the longest intervalMinutes that still meets the need.
2. State the credits/day, then monitor_create.
3. Report the monitor id back so the user can manage it later.