Price Watcher

Price Watcher

Track product prices from natural-language product descriptions. Do not require the user to provide product URLs. Store the watch query, discover and refresh multiple source URLs, save price history in SQLite, and write reports to the configured report directory.

Quick Start

Resolve paths relative to this skill directory. Use scripts/price_watcher.py for deterministic local state operations:

cd skills/price-watcher
python3 scripts/price_watcher.py init
python3 scripts/price_watcher.py add "Watch MacBook Pro M3 14-inch and notify me if it drops below $1200"
python3 scripts/price_watcher.py sources add --item-id 1 --site "Best Buy" --url "https://..."
python3 scripts/price_watcher.py record --item-id 1 --source-id 1 --price 1199.99 --currency USD
python3 scripts/price_watcher.py report

Default runtime paths:

• State root: ~/.codex/state/price-watcher/, override with PRICE_WATCHER_STATE
• Database: ~/.codex/state/price-watcher/price-watcher.sqlite3
• Reports: ~/.codex/state/price-watcher/price-reports/YYYY-MM-DD.md

Keep runtime databases and generated reports out of git unless the user explicitly asks to commit examples.

Workflow A: Add Watched Item

1. Parse the user's request into:
   • query: the product description to keep watching.
   • target_price: the numeric threshold if present.
   • currency: infer from symbols or wording; default to the user's locale when unclear.
2. Store the original product query, not a single fixed URL.
3. Search across multiple sources using web search, retailer search pages, or shopping/search APIs available in the current environment.
4. Normalize likely product matches:
   • canonical product name
   • brand/model/generation/size/storage/color or other important attributes
   • marketplace or retailer site
   • product URL
5. Deduplicate likely same-product matches by comparing normalized names and attributes. Prefer exact model/generation matches over merely similar products.
6. Add multiple source URLs for the item. Do not ask the user to manually provide URLs unless source discovery fails and the user wants to help.

Use python3 scripts/price_watcher.py add "<request>" after parsing to create the SQLite item, then add discovered source URLs with sources add.

Workflow B: Check Prices

For each active item:

1. Load active items and known sources from SQLite.
2. Refresh source discovery when:
   • no sources exist,
   • known sources fail repeatedly,
   • the user asks for broader coverage,
   • a product appears discontinued or out of stock everywhere.
3. Query all known sources. Use direct HTTP only when normal access works. Use Playwright/browser automation for pages that need client-side rendering.
4. Respect access controls:
   • Amazon scraping is allowed only through normal browser-accessible pages.
   • Do not bypass login, bot challenges, paywalls, CAPTCHAs, or account gates.
   • Do not purchase items, add to cart, reserve inventory, or initiate checkout.
5. Extract price, currency, availability when visible, and failure/skipped notes.
6. Normalize price and currency before storage. If conversion rates are needed and unavailable, keep the observed currency and do not compare across currencies as exact values.
7. Save each successful observation to price_checks.
8. Compare against the previous check for the same item/source and the item target price only when currencies are comparable.
9. Generate a Markdown report with report.

Report Requirements

Write reports to the configured price-reports/YYYY-MM-DD.md directory. Sort items lowest-price-first by the best current observed price. Include:

• item query and normalized name
• target price and whether the best price meets it
• source links
• observed prices and currencies
• lowest price across sources
• changes since the previous check
• failed or skipped source notes

Treat notifications as out of scope for v1. For a request like "notify me," store the threshold and write report status; do not send messages unless a later version adds a notification channel.

SQLite Schema

Use this schema exactly unless the user asks for an extension:

items(id, query, normalized_name, target_price, created_at, active)
sources(id, item_id, site, url, last_checked_at)
price_checks(id, item_id, source_id, price, currency, checked_at)

See references/schema.md for column meanings and operational notes.

Price Extraction

Prefer structured data in this order:

1. JSON-LD product offers
2. Open Graph/product meta tags
3. retailer page state JSON
4. visible DOM text near product title or buy box

Use references/source-discovery.md for source discovery and normalization heuristics. Use references/price-parsing.md for price parsing rules.

Failure Handling

Record failed or skipped sources in the report even when no price can be stored. Common notes:

• blocked: login required
• skipped: CAPTCHA or bot challenge
• skipped: checkout/cart-only price
• failed: price not visible
• failed: page unavailable

Do not invent prices. If a page shows a range, store the lowest visible purchasable price only when the report clearly states it was a range.