Proposal 001: complete a recurring catalog job
Proposed · 28 September 2026. This records the recommendation produced by the research task. It is not an approved implementation commitment.
Retain the merchant-facing, OpenCart-first direction and test reviewed supplier-price updates as the first paid workflow. Build the record of authorization and outcome before enabling writes.
The subsequent founder direction includes customer-facing agents alongside owner operations; see Market. This proposal still concerns the initial operations pilot and does not rule out that broader scope.
Existing direction retained
Store Deputy serves owners/operators, layers onto existing shop software, intends to use Agentfy, and begins with OpenCart. This repository remains documentation-only. No product functionality has been shipped by documenting it.
Recommended refinements
| Question | Recommendation | Reason |
|---|---|---|
| What to sell first? | A repeatable supplier-file-to-verified-price-update job | Concrete output, a visible baseline and repeated use to measure |
| What interface? | Chat plus Today, review tables and history | Intent and bulk verification need different surfaces |
| How much authority? | Read scoped data; draft freely; require explicit approval for the first writes | Makes the transaction and its consequences reviewable |
| When to build audit? | Before the first write | Recovery and authorization cannot be added after an uncertain mutation |
| How to integrate? | Versioned platform connector and deterministic execution | Business rules and retries must not depend on a model following a prompt |
| When to expand? | After paid repeat use and support economics are demonstrated | Multiple connectors multiply unknowns |
Alternatives considered
A generic admin chatbot is easier to explain but weakly differentiated. A reporting-only product could still be useful, but faces established alternatives. Shopper service has a different workflow and competitive set. Autonomous store management creates too much scope before trust and data quality are known. A source-grounded catalog-content workflow remains the leading alternative if interviews reject price maintenance.
Consequences and unresolved decisions
We accept narrower launch scope and a substantial connector/recovery burden. We have not yet chosen a release family, geography, provider, paid offer or launch date. Founder acceptance and pilot evidence should update this record explicitly. Implementation is not authorized by this proposal itself.
Supporting material: market, MVP, architecture, validation.