winston-kb: an internal knowledge base Skill across 188 sourced documents.

What it does.

Ask a question in plain language. The Skill embeds the query, queries the Supabase vector store across 146 reference documents and 42 standard operating procedures, and returns a sourced answer with citations to the underlying doc. No more hunting through Drive folders or pinging the senior strategist on Slack for the third time this week.

How we use it.

Anyone on the team can trigger it from a Claude conversation. New strategists onboard against the same knowledge the founder is working from. Client-facing answers stay consistent across engagements because everyone is pulling from the same source library. The Skill cites the doc, so the user can verify and dig deeper if needed.

Why we built it.

Internal knowledge dies in two places: long-tenure heads and unsearchable Drive folders. Both got expensive. A Retrieval Augmented Generation (RAG) Skill keeps the knowledge alive, queryable from the same chat where the work happens, and grows with every doc and SOP added to the corpus. Claude reads the source, the user reads the answer, the doc gets cited.

The stack inside it.

• Supabase Postgres with the pgvector extension as the vector store
• Anthropic API for embeddings and final answer generation
• Claude Skills as the chat-side activation layer
• 146 reference documents covering SEO, GEO, technical SEO, local SEO, content, links, schema
• 42 standard operating procedures covering audit, content, and reporting workflows
• A small ingestion pipeline that chunks, embeds, and upserts new docs into Supabase

What this proves.

Skills compound. Every job that was a half-day or a multi-day stretch of senior time becomes a Skill the team triggers from chat. The output ships faster. The strategist's calendar opens up for the work that actually needs human judgment: positioning, narrative, the prioritization call on what the data is really saying. That is the Winston operating model in one Skill.