campfirein/byterover-plugin/byterover-company-brain

Use the Company Brain

1. Call brain__query before answering a company-specific question. If the host defers MCP tools, use tool discovery to load and then invoke brain__query; finding the tool is not a query result.
2. For a company brief or broad question, call brain__query with {} first. This overview names readable Brains and the Brain that answered. Follow nextCursor with { cursor } for at most three pages. If hasMore is true but nextCursor is null, say that stable continuation is unavailable.
3. For a focused lookup, use one concise, distinctive name or concept. Use budget: "small" for a narrow answer, "medium" by default, and "large" only for broad context. Do not copy a long user question or combine unrelated categories into one query.
4. Use brain__query({ objectId }) when you need the complete accepted object named by an earlier result. Do not combine objectId with q, type, limit, cursor, or operationId; brain and budget may still select its source and envelope.
5. Send brain only when the user names a specific readable Brain. Use its exact id or name from the overview. Never guess a Brain id.
6. Base claims attributed to the Company Brain only on returned objects. Distinguish missing context from a contradictory fact. Never fill a gap by guessing.
7. Include useful source labels and the returned Brain revision when they help the reader verify the answer. If brain__query is unavailable after tool discovery, report that ByteRover Brain is not loaded in this agent session. If brain__query or brain__learn returns HTTP 401, invalid_token, or invalid_grant, report that ByteRover Brain is not authenticated, use the host's MCP login flow, and retry. Never ask for Company IDs, credentials, API endpoints, or local paths.
8. brain__query reads accepted context only. Do not claim that a missing statement is accepted Brain knowledge.
9. Use trigger: "user" only when the user explicitly asks to save that exact statement. For knowledge detected by the agent, use trigger: "agent", include a short safe context, and include contextOperationId when the statement came from an earlier Brain query. Never infer or expand the statement.
10. Claim that the Brain changed only when the outcome is added or updated. For queued, use brain__query({ operationId }) later to check status. For any other outcome, say that the statement was not saved and include the returned reason. If the tool returns insufficient_scope, reconnect and grant brain:learn.