From dc303222acf9ecb092f3f07b20ef38c5bb117af7 Mon Sep 17 00:00:00 2001
From: ciregenz <ciregenz@berkeley.edu>
Date: Sun, 16 Aug 2026 01:53:28 -0700
Subject: [PATCH] [eric] skills: frontmatter parser reads YAML block scalars,
 so a |- description no longer renders literally (ENG-307)

---
 .../skill_registry/skill_registry_github.py   | 22 +++++++++---
 backend/apps/skills/skills.py                 | 14 ++------
 .../tests/test_frontmatter_block_scalars.py   | 35 +++++++++++++++++++
 3 files changed, 55 insertions(+), 16 deletions(-)
 create mode 100644 backend/tests/test_frontmatter_block_scalars.py

diff --git a/backend/apps/skill_registry/skill_registry_github.py b/backend/apps/skill_registry/skill_registry_github.py
index d867c70a..4a560647 100644
--- a/backend/apps/skill_registry/skill_registry_github.py
+++ b/backend/apps/skill_registry/skill_registry_github.py
@@ -23,10 +23,24 @@ def parse_frontmatter(raw: str) -> tuple[dict, str]:
     fm_block = raw[3:end].strip()
     body = raw[end + 3:].strip()
     meta: dict = {}
-    for line in fm_block.splitlines():
-        m = re.match(r"^(\w[\w_-]*)\s*:\s*(.+)$", line)
-        if m:
-            meta[m.group(1).strip()] = m.group(2).strip().strip('"').strip("'")
+    lines = fm_block.splitlines()
+    i = 0
+    while i < len(lines):
+        m = re.match(r"^(\w[\w_-]*)\s*:\s*(.*)$", lines[i])
+        i += 1
+        if not m:
+            continue
+        key, val = m.group(1).strip(), m.group(2).strip()
+        if re.fullmatch(r"[|>][+-]?", val):
+            # YAML block scalar: the text is the indented lines below; the bare indicator once shipped as a card's whole description (ENG-307).
+            block: list[str] = []
+            while i < len(lines) and (not lines[i].strip() or lines[i][0] in (" ", "\t")):
+                block.append(lines[i].strip())
+                i += 1
+            joiner = "\n" if val[0] == "|" else " "
+            meta[key] = joiner.join(b for b in block if b).strip()
+        else:
+            meta[key] = val.strip('"').strip("'")
     return meta, body
 
 
diff --git a/backend/apps/skills/skills.py b/backend/apps/skills/skills.py
index 878252fe..70572b95 100644
--- a/backend/apps/skills/skills.py
+++ b/backend/apps/skills/skills.py
@@ -12,6 +12,7 @@ import time
 import zipfile
 from contextlib import asynccontextmanager
 from fastapi import HTTPException
+from backend.apps.skill_registry.skill_registry_github import parse_frontmatter
 from backend.config.Apps import SubApp
 from backend.apps.skills.models import Skill, SkillCreate, SkillLoadRequest, SkillUpdate, SkillUpload, SkillWorkspaceSeedRequest
 
@@ -339,18 +340,7 @@ async def load_skill(body: SkillLoadRequest):
 
 def p_parse_skill_frontmatter(raw: str) -> dict:
     """Extract YAML frontmatter fields from a SKILL.md file."""
-    if not raw.startswith("---"):
-        return {}
-    end = raw.find("---", 3)
-    if end == -1:
-        return {}
-    fm_block = raw[3:end].strip()
-    meta: dict = {}
-    for line in fm_block.splitlines():
-        m = re.match(r"^(\w[\w_-]*)\s*:\s*(.+)$", line)
-        if m:
-            meta[m.group(1).strip()] = m.group(2).strip().strip('"').strip("'")
-    return meta
+    return parse_frontmatter(raw)[0]
 
 
 @skills.router.post("/workspace/seed")
diff --git a/backend/tests/test_frontmatter_block_scalars.py b/backend/tests/test_frontmatter_block_scalars.py
new file mode 100644
index 00000000..1f4f9b36
--- /dev/null
+++ b/backend/tests/test_frontmatter_block_scalars.py
@@ -0,0 +1,35 @@
+"""A SKILL.md whose description uses a YAML block scalar rendered as the literal string "|-" on
+its marketplace card (ENG-307, seen live on /claude-api). The line-regex parser matched the
+indicator as the value and dropped the indented block below it. One parser now serves both the
+registry and the upload path, and it reads block scalars.
+"""
+from backend.apps.skill_registry.skill_registry_github import parse_frontmatter
+from backend.apps.skills.skills import p_parse_skill_frontmatter
+
+
+def test_literal_block_scalar_reads_the_indented_text():
+    meta, body = parse_frontmatter("---\nname: claude-api\ndescription: |-\n  Talk to the API\n  with retries.\n---\nBody")
+    assert meta["description"] == "Talk to the API\nwith retries."
+    assert meta["name"] == "claude-api"
+    assert body == "Body"
+
+
+def test_folded_scalar_joins_with_spaces():
+    meta, _ = parse_frontmatter("---\ndescription: >-\n  One line\n  folded.\n---\n")
+    assert meta["description"] == "One line folded."
+
+
+def test_plain_and_quoted_values_are_unchanged():
+    meta, _ = parse_frontmatter('---\nname: "quoted"\nlicense: MIT\n---\n')
+    assert meta == {"name": "quoted", "license": "MIT"}
+
+
+def test_empty_block_scalar_yields_empty_not_the_indicator():
+    meta, _ = parse_frontmatter("---\ndescription: |-\nname: x\n---\n")
+    assert meta["description"] == ""
+    assert meta["name"] == "x"
+
+
+def test_upload_path_uses_the_same_parser():
+    meta = p_parse_skill_frontmatter("---\ndescription: |-\n  Uploaded skill.\n---\n")
+    assert meta["description"] == "Uploaded skill."