Skip to content

[Community Submission] Add skill: map-google-product-taxonomy#88

Open
raqibdev wants to merge 6 commits intotimctfl:mainfrom
raqibdev:map-products-to-google-product-taxonomy
Open

[Community Submission] Add skill: map-google-product-taxonomy#88
raqibdev wants to merge 6 commits intotimctfl:mainfrom
raqibdev:map-products-to-google-product-taxonomy

Conversation

@raqibdev
Copy link
Copy Markdown
Contributor

Description

Add: Map Products to Google Product Taxonomy skill

Maps product CSVs to Google Product Taxonomy categories and outputs a supplemental feed with google_product_category filled in, ready to merge into a Google Shopping feed.


What's included

  • SKILL.md — full skill prompt with conversation flow, classification rules, and output spec
  • skillshelf.yaml — skill metadata and FAQ
  • scripts/classify_taxonomy.py — keyword classifier (Python 3.10+), outputs structured JSON with confidence tiers
  • assets/taxonomy-keywords.json — keyword index covering 93 categories across 15 verticals
  • references/ — example output, JSON schema spec, and glossary

Key design decisions

Decision Rationale
Hybrid script + LLM approach Deterministic keyword matching handles the majority of catalog items fast and cheaply. LLM handles ambiguous and low-signal cases that keyword matching cannot resolve.
Three confidence tiers (high / medium / low) Gives the merchant a clear action for each group: batch-approve, confirm individually, or review. Avoids presenting all results the same way regardless of certainty.
Policy flags on the result object Alcohol miscategorisation and missing apparel attributes are account-suspension risks in Google Shopping. Encoding them as flags on each result — rather than a post-hoc check — ensures they are surfaced before output is written.
--preserve-existing flag Merchants re-running the skill on a partially-mapped feed should not have existing 3-level+ assignments overwritten. The flag makes incremental runs safe by default.
DEPTH_BONUS scoring Deeper taxonomy paths win ties without overriding genuine score gaps. Capped at 0.5 so no realistic keyword match is overridden by depth alone.
LLM fallback is explicit, not silent If the script fails, the skill notes it in the report and attributes all classifications to LLM. Merchants can audit the source column and know what to trust.
Bundle classification rule Bundles are classified by primary item, not accessories or packaging. The is_bundle flag on the result object makes this auditable rather than implicit.

image image image image

Contributor Attestation

  • I agree to the Terms of Service and Privacy Policy, and I understand that if my submission is accepted it will be published under the Apache 2.0 license. I confirm I have the rights to submit this content and it does not include confidential information or personal data.

…orting features

- Implemented `classify_taxonomy.py` for classifying products against Google Product Taxonomy using keyword matching.
- Created example output documentation for mapping products to Google Product Taxonomy.
- Added glossary for supplemental feed CSV output to clarify field definitions and usage.
- Defined YAML configuration for the new skill, including metadata and author information.
…dle detection, and incremental processing support.
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant