mempalace.auto_query.formatter
Source: mempalace/auto_query/formatter.py
Result formatter for the auto-query integration.
Takes an MCP tool call result and formats it for injection into the Claude Code system message. Output is verbatim snippets from palace drawers, wrapped in sentinel tokens for reliable detection by feedback collectors.
The formatter handles four MCP result shapes:
- mempalace_search (semantic/hybrid search results)
- mempalace_kg_query (knowledge graph entity relationships)
- mempalace_diary_read (per-agent diary entries)
- mempalace_traverse (cross-wing graph traversal)
Functions
format_injection
python
def format_injection(tool_call: MCPCall, mcp_result: dict, signals: SignalSet, latency_ms: int) -> Optional[str]Format MCP results as an injection block with sentinel tokens.
Returns the formatted injection string, or None if no results found. The output is verbatim -- no summarisation. Each result is a truncated preview of the raw drawer/entry content.
