fix: preserve folded skill descriptions in s05

agents/s05_skill_loading.py

@@ -76,12 +76,44 @@ def _parse_frontmatter(self, text: str) -> tuple:
         if not match:
             return {}, text
         meta = {}
-        for line in match.group(1).strip().splitlines():
-            if ":" in line:
-                key, val = line.split(":", 1)
-                meta[key.strip()] = val.strip()
+        lines = match.group(1).strip().splitlines()
+        index = 0
+        while index < len(lines):
+            line = lines[index]
+            if ":" not in line:
+                index += 1
+                continue
+
+            key, val = line.split(":", 1)
+            key = key.strip()
+            val = val.strip()
+            index += 1
+
+            if val in {"|", ">"}:
+                block_lines = []
+                while index < len(lines):
+                    next_line = lines[index]
+                    if next_line.startswith((" ", "\t")) or not next_line.strip():
+                        block_lines.append(next_line)
+                        index += 1
+                        continue
+                    break
+                meta[key] = self._parse_block_scalar(block_lines, val)
+                continue
+
+            meta[key] = val
         return meta, match.group(2).strip()
 
+    def _parse_block_scalar(self, lines: list[str], style: str) -> str:
+        """Parse a minimal YAML block scalar for skill metadata."""
+        non_empty = [len(line) - len(line.lstrip()) for line in lines if line.strip()]
+        indent = min(non_empty) if non_empty else 0
+        normalized = [line[indent:] if line.strip() else "" for line in lines]
+
+        if style == ">":
+            return " ".join(line.strip() for line in normalized if line.strip())
+        return "\n".join(normalized).strip()
+
     def get_descriptions(self) -> str:
         """Layer 1: short descriptions for the system prompt."""
         if not self.skills:

tests/test_s05_skill_loading.py

@@ -0,0 +1,90 @@
+import importlib.util
+import os
+import sys
+import tempfile
+import textwrap
+import types
+import unittest
+from pathlib import Path
+
+
+REPO_ROOT = Path(__file__).resolve().parents[1]
+MODULE_PATH = REPO_ROOT / "agents" / "s05_skill_loading.py"
+
+
+def load_s05_module(temp_cwd: Path):
+    fake_anthropic = types.ModuleType("anthropic")
+
+    class FakeAnthropic:
+        def __init__(self, *args, **kwargs):
+            self.messages = types.SimpleNamespace(create=None)
+
+    fake_dotenv = types.ModuleType("dotenv")
+    setattr(fake_anthropic, "Anthropic", FakeAnthropic)
+    setattr(fake_dotenv, "load_dotenv", lambda override=True: None)
+
+    previous_anthropic = sys.modules.get("anthropic")
+    previous_dotenv = sys.modules.get("dotenv")
+    previous_cwd = Path.cwd()
+    spec = importlib.util.spec_from_file_location("s05_under_test", MODULE_PATH)
+    if spec is None or spec.loader is None:
+        raise RuntimeError(f"Unable to load {MODULE_PATH}")
+    module = importlib.util.module_from_spec(spec)
+
+    sys.modules["anthropic"] = fake_anthropic
+    sys.modules["dotenv"] = fake_dotenv
+    try:
+        os.chdir(temp_cwd)
+        os.environ.setdefault("MODEL_ID", "test-model")
+        spec.loader.exec_module(module)
+        return module
+    finally:
+        os.chdir(previous_cwd)
+        if previous_anthropic is None:
+            sys.modules.pop("anthropic", None)
+        else:
+            sys.modules["anthropic"] = previous_anthropic
+        if previous_dotenv is None:
+            sys.modules.pop("dotenv", None)
+        else:
+            sys.modules["dotenv"] = previous_dotenv
+
+
+class SkillLoaderTests(unittest.TestCase):
+    def test_get_descriptions_keeps_folded_frontmatter_description(self):
+        with tempfile.TemporaryDirectory() as tmp:
+            temp_cwd = Path(tmp)
+            module = load_s05_module(temp_cwd)
+            skills_dir = temp_cwd / "skills" / "demo-skill"
+            skills_dir.mkdir(parents=True)
+            (skills_dir / "SKILL.md").write_text(
+                textwrap.dedent(
+                    """\
+                    ---
+                    name: demo
+                    description: >
+                      Loads the skill metadata correctly
+                      even when the description spans lines.
+                    tags: parsing, regression
+                    ---
+                    # Demo Skill
+
+                    Body content.
+                    """
+                )
+            )
+
+            loader = module.SkillLoader(temp_cwd / "skills")
+
+            self.assertEqual(
+                loader.skills["demo"]["meta"]["description"],
+                "Loads the skill metadata correctly even when the description spans lines.",
+            )
+            self.assertEqual(
+                loader.get_descriptions(),
+                "  - demo: Loads the skill metadata correctly even when the description spans lines. [parsing, regression]",
+            )
+
+
+if __name__ == "__main__":
+    unittest.main()