fix(schema-pack): reject YAML block scalars loudly instead of silently truncating manifests (#1750)

Takeover of #1799 (fork branch rotted against master; refreshed CI hit an
unrelated shard-timeout flake). parseYamlMini now throws a clear error when
a bare block-scalar indicator (|, >, |-, >+, |2) reaches parseScalar,
instead of mis-parsing it as the literal string and silently dropping every
following top-level key. Inlines gbrain-recommended.yaml's description so
the bundled pack loads its full page_types.

Co-authored-by: harjoth <harjoth.khara@gmail.com>
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>

src/core/schema-pack/base/gbrain-recommended.yaml

@@ -26,12 +26,7 @@ name: gbrain-recommended
 version: 1.0.0
 gbrain_min_version: 0.39.0
 extends: gbrain-base
-description: |
-  The recommended starter pack for an operational personal brain.
-  Based on docs/GBRAIN_RECOMMENDED_SCHEMA.md — Karpathy's LLM wiki
-  pattern extended into a full operational knowledge base. Use this
-  when you want the documented behavior immediately instead of
-  authoring your own pack from scratch.
+description: "The recommended starter pack for an operational personal brain. Based on docs/GBRAIN_RECOMMENDED_SCHEMA.md — Karpathy's LLM wiki pattern extended into a full operational knowledge base. Use this when you want the documented behavior immediately instead of authoring your own pack from scratch."
 
 # Pack-specific page_types (extend gbrain-base's seed types).
 # Each maps a path prefix to a primitive + opt-in flags. Most types

src/core/schema-pack/loader.ts

@@ -84,7 +84,9 @@ export function loadPackFromString(content: string, hint: string): SchemaPackMan
  *
  * Rejected by design: anchors (&), aliases (*), tags (!), flow style
  * ({...}, [...] except as JSON), block scalars (|, >), multi-document (---).
- * Pack authors who need these features should ship JSON.
+ * Pack authors who need these features should ship JSON. Block scalars in
+ * particular now throw a clear error (see parseScalar) rather than silently
+ * mis-parsing and truncating the rest of the manifest.
  *
  * This is intentionally narrow. The skill-pack and storage-config parsers
  * use similar hand-rolled patterns; this one is shape-customized for pack
@@ -133,6 +135,20 @@ export function parseYamlMini(content: string): unknown {
     // Number
     if (/^-?\d+$/.test(trimmed)) return parseInt(trimmed, 10);
     if (/^-?\d+\.\d+$/.test(trimmed)) return parseFloat(trimmed);
+    // Block scalars (`|`, `>`, with optional chomping/indent indicators such
+    // as `|-`, `>+`, `|2`) are unsupported by design — see the header. A bare
+    // indicator reaching here is a `key: |` line whose indented body would
+    // otherwise silently truncate the rest of the mapping: parseScalar would
+    // return the literal "|"/">", and the over-indented body would trip the
+    // `indent > baseIndent` break in parseMapping, dropping every following
+    // top-level key with no diagnostic. Fail loud instead of corrupting.
+    if (/^[|>][0-9+-]*$/.test(trimmed)) {
+      throw new Error(
+        `parseYamlMini: block scalars are not supported (found "${trimmed}"). ` +
+          `Use an inline or quoted string (e.g. description: "..."), or ship the manifest as JSON. ` +
+          `Left unhandled, a block scalar silently drops every key after it in the manifest.`,
+      );
+    }
     // Bare string
     return trimmed;
   }

test/schema-pack-loader.test.ts

@@ -10,6 +10,7 @@
 // module's internal contracts.
 
 import { describe, expect, test } from 'bun:test';
+import { readFileSync } from 'node:fs';
 import { withEnv } from './helpers/with-env.ts';
 import {
   buildAliasGraph,
@@ -349,6 +350,38 @@ describe('YAML mini-parser', () => {
     const result = parseYamlMini('# top comment\nname: value # inline comment') as Record<string, unknown>;
     expect(result.name).toBe('value');
   });
+
+  // Regression: block scalars (`|`, `>`) are unsupported by design. Before
+  // the fix, `description: |` was read as the literal "|" and its indented
+  // body silently truncated the mapping — every key after it was dropped
+  // with no error (issue #1750). Now it fails loud.
+  test('throws a clear error on block scalars instead of silently truncating', () => {
+    expect(() => parseYamlMini('description: |\n  multi\n  line\npage_types:\n  - name: deal'))
+      .toThrow(/block scalars are not supported/);
+    expect(() => parseYamlMini('description: >\n  folded text')).toThrow(/block scalars/);
+    // Chomping / indent indicators (`|-`, `>+`, `|2`) are caught too.
+    expect(() => parseYamlMini('x: |-\n  body')).toThrow(/block scalars/);
+    expect(() => parseYamlMini('x: |2\n  body')).toThrow(/block scalars/);
+  });
+
+  test('a quoted "|" is still a normal string (not mistaken for a block scalar)', () => {
+    const result = parseYamlMini('sep: "|"') as Record<string, unknown>;
+    expect(result.sep).toBe('|');
+  });
+
+  // The bundled gbrain-recommended pack uses a (now inline) description and
+  // declares 12 page_types. Before #1750 it silently resolved to 0 types.
+  test('bundled gbrain-recommended.yaml parses with its full page_types', () => {
+    const yaml = readFileSync(
+      `${import.meta.dir}/../src/core/schema-pack/base/gbrain-recommended.yaml`,
+      'utf8',
+    );
+    const parsed = parseYamlMini(yaml) as { page_types?: Array<{ name: string }>; description?: string };
+    expect(Array.isArray(parsed.page_types)).toBe(true);
+    expect(parsed.page_types!.length).toBeGreaterThanOrEqual(12);
+    expect(parsed.page_types!.map(t => t.name)).toContain('deal');
+    expect(parsed.description).not.toBe('|');
+  });
 });
 
 describe('loadPackFromString end-to-end', () => {