Content Cleaner

the-brain provides built-in content cleaning utilities that strip noise from harvested interactions.

Functions

cleanMemoryContent(content: string): string

Strips Claude XML artifacts and extracts meaningful user requests from observations.

What it removes:

• <function_calls> XML blocks
• <tool_call> blocks
• <what_happened> wrappers
• observed_from_primary_session markers
• Progress summary boilerplate
• Claude-Mem observer preambles

What it preserves:

• User requests embedded in observations
• Bash commands from Bash observations
• File paths from Edit observations
• MCP tool calls

Example:

import { cleanMemoryContent } from "@the-brain/core";

const raw = `<what_happened>User asked about React hooks</what_happened>
<function_calls><tool_call>read_file</tool_call></function_calls>`;

const cleaned = cleanMemoryContent(raw);
// → "User asked about React hooks"

cleanGraphNodeLabel(label: string): string

Trims long code-fragment labels to keep graph nodes concise. Uses Unicode-aware sentence splitting (., !, ?, 。, ！, ？) to work across all languages.

const label = cleanGraphNodeLabel("use `useCallback` instead of `useMemo` for the handler function in the parent component");
// → "use `useCallback` instead of `useMemo`"

deduplicateContents(items: Array<{content: string}>): Array

Keeps the highest-signal version when duplicate content is detected.

const deduped = deduplicateContents([
  { content: "Fix auth bug" },
  { content: "Fix auth bug in login component" },
]);
// → [{ content: "Fix auth bug in login component" }]  (longer = higher signal)