feat(core): implement parseMemory with structured section extraction

platform/core/src/index.ts

@@ -114,7 +114,7 @@ export {
 export type { ModelCatalogProvider, PipelineModelPreset } from "./llm-model-catalog";
 export { parseManifest, generateManifest } from "./manifest";
 export { parseSoul, generateSoul } from "./soul";
-export { parseMemory, generateMemory } from "./memory";
+export { parseMemory, generateMemory, type ParsedMemory } from "./memory";
 export {
 	NETWORK_MODES,
 	normalizeNetworkMode,

platform/core/src/memory.ts

@@ -1,6 +1,61 @@
-export function parseMemory(markdown: string): Record<string, any> {
-	// TODO: Parse memory markdown into structured data
-	return { raw: markdown };
+export interface ParsedMemory {
+	/** Content under the "## User Profile" section. */
+	userProfile: string;
+	/** Content under the "## Key Facts" section. */
+	keyFacts: string;
+	/** Content under the "## Ongoing Context" section. */
+	ongoingContext: string;
+	/** Content between MANUAL:START and MANUAL:END markers. */
+	manualNotes: string;
+	/** The full raw markdown input, preserved for round-tripping. */
+	raw: string;
+}
+
+/**
+ * Parse a Signet memory markdown file into structured sections.
+ *
+ * Extracts content from well-known `## ` headings and the
+ * `<!-- MANUAL:START -->` / `<!-- MANUAL:END -->` block. Any content
+ * outside recognized sections is ignored — the `raw` field always
+ * contains the original markdown for lossless round-tripping.
+ */
+export function parseMemory(markdown: string): ParsedMemory {
+	const sections: Record<string, string> = {};
+	let currentSection: string | null = null;
+	const sectionLines: string[] = [];
+
+	const lines = markdown.split("\n");
+	for (const line of lines) {
+		const headingMatch = line.match(/^##\s+(.+)/);
+		if (headingMatch) {
+			// Flush previous section
+			if (currentSection !== null) {
+				sections[currentSection] = sectionLines.join("\n").trim();
+			}
+			currentSection = headingMatch[1].trim();
+			sectionLines.length = 0;
+		} else if (currentSection !== null) {
+			sectionLines.push(line);
+		}
+	}
+	// Flush last section
+	if (currentSection !== null) {
+		sections[currentSection] = sectionLines.join("\n").trim();
+	}
+
+	// Extract manual notes block
+	const manualMatch = markdown.match(
+		/<!--\s*MANUAL:START\s*-->([\s\S]*?)<!--\s*MANUAL:END\s*-->/,
+	);
+	const manualNotes = manualMatch ? manualMatch[1].trim() : "";
+
+	return {
+		userProfile: sections["User Profile"] ?? "",
+		keyFacts: sections["Key Facts"] ?? "",
+		ongoingContext: sections["Ongoing Context"] ?? "",
+		manualNotes,
+		raw: markdown,
+	};
 }
 
 export function generateMemory(): string {