Skip to project details
eBay Listing Tool
Menu

Listing automation

eBay Listing Tool

A private local-first system intended to replace Spadeberry while keeping personal photos and processing on owner-controlled infrastructure. Its core pipeline, review workbench, and deployment contract are built; Nifty remains the separate cross-listing tool, and production publishing remains disabled.

Private working systemPrivate local-first workflow; production publishing remains disabled pending sandbox proof and approval

Local listing workbench

Private local system · production publishing disabled
Review GridPrivacy-safe representative state
AllReviewDraftApprovedVerifiedFlagged
PHOTOWITHHELDLocal source
LOCAL DRAFT · ITEM WITHHELDListing details stay private until review.
SIZE UNCONFIRMEDHUMAN REVIEW
Required evidenceSize not confirmedHuman confirmation is mandatory
ConditionReview requiredPricing withheld
PUBLISH LOCKEDResolve required fields
01

The problem

Preparing consistent clothing listings requires repeated photo review, tag reading, standardized fields, title construction, pricing checks, condition judgment, and post-publish verification. Those steps need to be dependable without allowing uncertain data or an approval shortcut to publish a listing.

02

The current approach

The system uses a locked local vision model, standardized clothing attributes, deterministic title and description assembly, sold-history and fee calculations, an editable review grid, and explicit publish-and-verify controls.

A local review path

Folder-based intake and processing feed the built review workbench, where required data and a human-confirmed standardized size gate approval. Per-item and batch publishing and verification are implemented, but a real eBay sandbox publish-retrieve-verify run still requires developer access, credentials, configuration, safe fixture URLs, and explicit authorization.

  • Local photo intake, gemma3:12b structured vision, and bounded clothing attributes
  • Deterministic titles, modular descriptions, fee math, and imported sold-history support
  • Editable review grid with mandatory size evidence, explicit size confirmation, and condition gates
  • Batch publish and verification results plus a Docker and Unraid runtime contract

Safeguards & boundaries

Every standardized size requires explicit human confirmation; anything below Excellent, any possible damage, missing costs, or uncertain required data enters manual review.

Personal photos remain local, Nifty remains the separate cross-listing and per-platform pricing tool, and production publishing stays disabled until sandbox proof succeeds and a separate rollout is approved.

What supports the status today

Private local-first workflow; production publishing remains disabled pending sandbox proof and approval

  • The current repository passes 223 automated tests with two credential-gated sandbox tests skipped, and Ruff reports no lint errors.
  • The completed 51-item evaluation records fresh unified two-pass results: gemma3:12b size 49/51 and item type 44/51; gemma3:27b size 47/51 and item type 46/51; the field-wise combination (12b size + 27b item type) gets both priority fields right on 45/51 (88.2%); measurement abstention is 51/51 for both.
  • The core pipeline, review workbench, publish-and-verify results, media upload, and Docker/Unraid runtime contract are built, but no real sandbox publishing success is claimed.
Current outcome

A working private local-first system whose core pipeline, review workbench, and deployment contract are built while production publishing remains sandbox-gated.

Next step

Maintain the completed 51-item evaluation record, obtain the required sandbox access and configuration, record an authorized end-to-end sandbox run, and seek separate approval before production rollout.

Supporting technology
  • Python
  • FastAPI
  • SQLite
  • Ollama
  • eBay APIs
  • Docker
  • pytest
  • Ruff