From 40d4c13c4da49df801651c9424abcaf8a43f03d9 Mon Sep 17 00:00:00 2001
From: Yigtwxx <yigiterdogan023@gmail.com>
Date: Fri, 17 Jul 2026 23:39:23 +0300
Subject: [PATCH] fix(skills): load ray-data and ray-train by quoting YAML
 extras

Both SKILL.md files declared dependencies as [ray[data], ...] and
[ray[train], ...]. In a YAML flow sequence the inner [ opens a nested
sequence, so gray-matter/js-yaml cannot parse the frontmatter. The skill
loader catches the parse failure and drops both skills, so they never
appear in `skill list`, the / picker, or the Skills panel, with no
user-visible error. Quote the extras so each entry is a plain scalar.

Add a test that parses every bundled SKILL.md and validates its skill
frontmatter, so a malformed block fails CI instead of silently dropping
a skill.

Fixes #187
---
 .../skills/data-engineering/ray-data/SKILL.md |  2 +-
 .../cli/skills/ml-training/ray-train/SKILL.md |  2 +-
 backend/cli/test/skill/bundled-skills.test.ts | 40 +++++++++++++++++++
 3 files changed, 42 insertions(+), 2 deletions(-)
 create mode 100644 backend/cli/test/skill/bundled-skills.test.ts

diff --git a/backend/cli/skills/data-engineering/ray-data/SKILL.md b/backend/cli/skills/data-engineering/ray-data/SKILL.md
index 6365e87d..3098a515 100644
--- a/backend/cli/skills/data-engineering/ray-data/SKILL.md
+++ b/backend/cli/skills/data-engineering/ray-data/SKILL.md
@@ -6,7 +6,7 @@ version: 1.0.0
 author: Synthetic Sciences
 license: MIT
 tags: [Data Processing, Ray Data, Distributed Computing, ML Pipelines, Batch Inference, ETL, Scalable, Ray, PyTorch, TensorFlow]
-dependencies: [ray[data], pyarrow, pandas]
+dependencies: ["ray[data]", pyarrow, pandas]
 ---
 
 # Ray Data - Scalable ML Data Processing
diff --git a/backend/cli/skills/ml-training/ray-train/SKILL.md b/backend/cli/skills/ml-training/ray-train/SKILL.md
index 2770e49e..d2a1590d 100644
--- a/backend/cli/skills/ml-training/ray-train/SKILL.md
+++ b/backend/cli/skills/ml-training/ray-train/SKILL.md
@@ -6,7 +6,7 @@ version: 1.0.0
 author: Synthetic Sciences
 license: MIT
 tags: [Ray Train, Distributed Training, Synthetic Sciencestion, Ray, Hyperparameter Tuning, Fault Tolerance, Elastic Scaling, Multi-Node, PyTorch, TensorFlow]
-dependencies: [ray[train], torch, transformers]
+dependencies: ["ray[train]", torch, transformers]
 ---
 
 # Ray Train - Distributed Training Synthetic Sciencestion
diff --git a/backend/cli/test/skill/bundled-skills.test.ts b/backend/cli/test/skill/bundled-skills.test.ts
new file mode 100644
index 00000000..21a36847
--- /dev/null
+++ b/backend/cli/test/skill/bundled-skills.test.ts
@@ -0,0 +1,40 @@
+import { test, expect } from "bun:test"
+import { ConfigMarkdown } from "../../src/config/markdown"
+import { Skill } from "../../src/skill"
+import path from "path"
+
+// Same fields the loader validates in Skill.compute() before accepting a skill.
+const Frontmatter = Skill.Info.pick({ name: true, description: true, category: true, tags: true, entry: true })
+
+const root = path.join(import.meta.dir, "..", "..", "skills")
+const files = await Array.fromAsync(new Bun.Glob("**/SKILL.md").scan({ cwd: root, absolute: true }))
+
+// Every bundled SKILL.md that declares a frontmatter block must produce a
+// loadable skill. Malformed YAML (e.g. the unquoted `ray[train]` extras in
+// issue #187) leaves the parser with empty data, so the loader silently drops
+// the skill and it never appears in `skill list`. Category READMEs carry no
+// frontmatter block and are intentionally skipped.
+test("every bundled skill with frontmatter loads", async () => {
+  expect(files.length).toBeGreaterThan(0)
+  const broken = await Promise.all(
+    files.map(async (file) => {
+      const raw = await Bun.file(file).text()
+      if (!raw.startsWith("---")) return undefined
+      const parsed = await ConfigMarkdown.parse(file)
+      return Frontmatter.safeParse(parsed.data).success ? undefined : path.relative(root, file)
+    }),
+  )
+  expect(broken.filter(Boolean)).toEqual([])
+})
+
+test("ray-train frontmatter parses with quoted extras", async () => {
+  const parsed = await ConfigMarkdown.parse(path.join(root, "ml-training", "ray-train", "SKILL.md"))
+  expect(parsed.data.name).toBe("ray-train")
+  expect(parsed.data.dependencies[0]).toBe("ray[train]")
+})
+
+test("ray-data frontmatter parses with quoted extras", async () => {
+  const parsed = await ConfigMarkdown.parse(path.join(root, "data-engineering", "ray-data", "SKILL.md"))
+  expect(parsed.data.name).toBe("ray-data")
+  expect(parsed.data.dependencies[0]).toBe("ray[data]")
+})