Ebay Seller

eBay-Seller

Guides each item through fixed phases. Per-item state lives in <workspace>/ebay/items/<YYYY-MM-DD-slug>/state.json. Seller defaults in <workspace>/ebay/config.json. Usually one new item per day; several items can run in parallel across different phases.

The listing text and item content are produced in German (target marketplace eBay.de), even though this guide is written in English.

<workspace> = the user's current project directory (fallback: ask). On Windows the .claude directory is under %USERPROFILE%\.claude\. References and templates live relative to this SKILL.md under references/ and templates/.

Requirements

• Required: Claude Code + the "Claude in Chrome" extension, and a Chrome where the user is logged in to eBay.de. Nothing else — runs on macOS and Windows.
• Optional (convenience only): Python 3 + Pillow for automatic photo enhancement (scripts/enhance_photo.py). Without Python/Pillow the skill still works — photos are used unedited (the user uploads them manually anyway).
• No installation needed: manage state.json directly (read/write JSON), no Python helper required.

Safety gates (non-negotiable)

• Never publish without explicit approval. Fill the listing, then stop before the final click.
• Live price changes and counter-offers only after approval.
• Never retouch defects out of photos (see references/compliance.md).

First run

If <workspace>/ebay/config.json does not exist, this is the first run: first show a short quick-start (3 sentences: drag in a photo + "sell on eBay"; you upload the photos yourself; nothing is published without your approval). Then create it from templates/config-template.json and ask the user for the fields (location/ZIP, shipping method+cost, returns, handling time, default sale format, tone). ebay_benutzername is optional. Login is manual: the user logs in to eBay in Chrome themselves — no password in chat, no Keychain setup needed.

Every run

1. Scan <workspace>/ebay/items/. Show an overview: item × phase × today's task.
2. New photo(s) on the message? (one or more) — derive a rough name from the first photo, then search items/ for an item with a similar name (dedupe). Match → ask whether it is the same item; if yes, add the new photos to the existing item. Otherwise create a new item (phase neu).
3. For items in beobachten and live: daily update — but only if letzter_lauf ≠ today (idempotency). Otherwise skip and report "already captured today".
4. The user drives the phase transitions.

State handling: read/write state.json directly. Save once at the end of a step; for daily updates set letzter_lauf to today's date in the same save. Never save mid browser step. (Optional, if Python is present: the helpers in scripts/state.py do the same.)

Note: phase values (neu, beobachten, fertig_machen, einstellen, live, abgeschlossen) and the JSON field names (fotos, kategorie, artikelmerkmale, preis_history, preisvorschlag, letzter_lauf, listing_url, live_history, …) are literal identifiers used by the templates and helpers — keep them exactly as written.

Phases

neu → identify

• See references/item-identification.md.
• Save all submitted photos under photos/original-N.jpg (1..n) and record them in fotos[] as {original} (first = main image).
• Fill artikel, kategorie, artikelmerkmale (as far as possible).
• First sold-price research (see references/price-research.md), write the first preis_history entry — only if letzter_lauf ≠ today (no double scraping).
• Phase → beobachten.

beobachten → build price sense

• Daily (idempotent) capture sold prices + competition, append to preis_history.
• Show the trend (median/range over the days).
• User says "ready" → phase fertig_machen.

fertig_machen → photos + price + text

• Photo enhancement (optional): if Python 3 + Pillow are present, enhance all photos — run from the skill folder: cd "<skill-folder>" && python3 -c "from scripts import enhance_photo; print(enhance_photo.enhance_all('<absolute-item-path>/photos'))" and write the returned enhanced names back into fotos[]. Without Python/Pillow, skip this step and use the original photos.
• Price suggestion from preis_history (median/range) with rationale → preisvorschlag. Range = fixed price resp. auction start price.
• Write the listing text from templates/listing-template.md → listing.md (see references/listing-guide.md). Write it in German.
• Optimization (see references/optimization.md): 80-character title with synonyms, main image = product, complete item specifics, format/best-offer strategy, optional free shipping, timing — present as suggestions to the user.
• User approval → phase einstellen.

einstellen → fill the listing

• Browser: check tab context, open a new tab. Make sure the user is logged in to eBay in Chrome — if not, ask them to log in (see references/login.md). Then the eBay listing form.
• Fields from state.json/config.json: title, description, category, item specifics, condition, images (existing fotos, first = main image), price/format, shipping profile, location/ZIP.
• The user uploads photos themselves (programmatic file upload is blocked for automation): drag the enhanced or original photos into the photo area.
• Ask the user for any missing required fields.
• STOP before the final publish. After approval, publish, save listing_url, phase → live.

live → monitor

• Daily (idempotent) read views/watchers/offers → live_history entry { datum, aufrufe, beobachter, angebote }.
• Anomalies (few views, many watchers without purchase) → SUGGEST a price/text adjustment, change only after approval.
• Sold/ended → phase abgeschlossen.

abgeschlossen

End state, folder kept for history.

Browser notes

• The eBay tools (Claude in Chrome) are deferred → load via ToolSearch before use (select:mcp__claude-in-chrome__...).
• Before any eBay action make sure the user is logged in; on 2FA/captcha, stop and ask the user.
• Work throttled (see references/price-research.md).