Context
TaxHacker already supports custom AI prompts for category extraction, which is great for per-user customization. But every user currently has to write their own classification logic from scratch — which tax categories to use, how to handle ambiguous transactions, what conservative defaults to apply.
Proposal
OpenAccountants is an open-source library of 371 tax classification skills covering 134 countries. Each skill contains:
- Supplier pattern libraries — e.g., "STRIPE PAYMENTS EU" → platform sales, "COMCAST BUSINESS" → utilities/deductible
- Tax category mappings — per-country VAT rates, income tax brackets, deduction rules
- Conservative default logic — when uncertain, assume more tax (not less), and flag the assumption
These could be adapted as pre-built prompt templates that TaxHacker users select by country instead of writing from scratch. For example, a US freelancer picks "US Schedule C" and gets classification rules for gross receipts, office expenses, utilities, meals (50% deductible), etc.
What this could look like
- A "Country templates" dropdown or config option in TaxHacker
- Each template maps to an OA skill file that gets injected into the LLM prompt
- Users can still customize on top of the template
Why this helps TaxHacker users
- No prompt engineering needed — pick your country, upload receipts, go
- Country-specific supplier patterns catch local merchants (Vodafone, Deutsche Telekom, Comcast, etc.)
- Conservative defaults reduce the chance of misclassification in the user's favor (which is the dangerous direction for tax)
Compatibility
- OA skills are MIT-friendly content (AGPL source, but the classification rules themselves are factual/statutory)
- TaxHacker is MIT licensed
- No dependency on OA infrastructure — the skill text would be embedded directly
Happy to contribute a proof-of-concept PR if there's interest. Would start with US Schedule C as the first template since it's the largest user base.
Repository: https://github.com/openaccountants/openaccountants
Context
TaxHacker already supports custom AI prompts for category extraction, which is great for per-user customization. But every user currently has to write their own classification logic from scratch — which tax categories to use, how to handle ambiguous transactions, what conservative defaults to apply.
Proposal
OpenAccountants is an open-source library of 371 tax classification skills covering 134 countries. Each skill contains:
These could be adapted as pre-built prompt templates that TaxHacker users select by country instead of writing from scratch. For example, a US freelancer picks "US Schedule C" and gets classification rules for gross receipts, office expenses, utilities, meals (50% deductible), etc.
What this could look like
Why this helps TaxHacker users
Compatibility
Happy to contribute a proof-of-concept PR if there's interest. Would start with US Schedule C as the first template since it's the largest user base.
Repository: https://github.com/openaccountants/openaccountants