sb-find

Stable Baseline: find and read

What this skill does

Answers questions from content held in Stable Baseline. It resolves the caller's scope first, then searches, reads and cites.

Everything runs under the signed in Cowork user's own Stable Baseline identity. There is no shared key and no service account in this package. If a user cannot see something in the Stable Baseline web app, they cannot see it here either.

IDs are never hardcoded. Every organisation, workspace, project, folder and document ID is resolved at runtime from the calling user's own token, on every request. Tools that require an ID, for example listProjects needing workspaceId or getDocument needing documentId, take it from a listing call made in this conversation. Never from memory, never from an earlier conversation.

Choosing the right tool

This is the most common source of wrong answers, so read it before searching.

To find an artefact by title, friendly ID such as DOC-123, or words in its body, use the listing tools. listDocuments is the main one. Despite the name it is a full content search: pass query and it greps across every document body, not just titles, returning contentMatches with line numbers and context. It accepts isRegex, caseSensitive, contextLines and maxMatchesPerDocument, and it returns versionTimestamp per document, so an edit can follow straight from the results with no getDocument round trip.

To find an artefact of ANY kind at once - document, whiteboard, plan, task, improvement or compliance item - use kg_search with artefactMetadataOnly: true. It matches title, friendly ID and UUID across every type and returns a typed list with an href for each hit. Narrow it with artefactTypes when you already know the kind. This mode reads no document bodies and does not need the knowledge graph, so it works on every tier.

To answer a question from the knowledge inside content, use kg_search WITHOUT that flag. That is the knowledge retriever: it reads inside content and will not reliably find something by its title.