fix: frontmatter was adding newlines in some cases causing invalid model ids

Co-authored-by: aptdnfapt <aptdnfapt@users.noreply.github.com>

packages/opencode/src/config/markdown.ts

@@ -14,7 +14,9 @@ export namespace ConfigMarkdown {
     return Array.from(template.matchAll(SHELL_REGEX))
   }
 
-  export function preprocessFrontmatter(content: string): string {
+  // other coding agents like claude code allow invalid yaml in their
+  // frontmatter, we need to fallback to a more permissive parser for those cases
+  export function fallbackSanitization(content: string): string {
     const match = content.match(/^---\r?\n([\s\S]*?)\r?\n---/)
     if (!match) return content
 
@@ -53,7 +55,7 @@ export namespace ConfigMarkdown {
 
       // if value contains a colon, convert to block scalar
       if (value.includes(":")) {
-        result.push(`${key}: |`)
+        result.push(`${key}: |-`)
         result.push(`  ${value}`)
         continue
       }
@@ -66,20 +68,23 @@ export namespace ConfigMarkdown {
   }
 
   export async function parse(filePath: string) {
-    const raw = await Bun.file(filePath).text()
-    const template = preprocessFrontmatter(raw)
+    const template = await Bun.file(filePath).text()
 
     try {
       const md = matter(template)
       return md
-    } catch (err) {
-      throw new FrontmatterError(
-        {
-          path: filePath,
-          message: `${filePath}: Failed to parse YAML frontmatter: ${err instanceof Error ? err.message : String(err)}`,
-        },
-        { cause: err },
-      )
+    } catch {
+      try {
+        return matter(fallbackSanitization(template))
+      } catch (err) {
+        throw new FrontmatterError(
+          {
+            path: filePath,
+            message: `${filePath}: Failed to parse YAML frontmatter: ${err instanceof Error ? err.message : String(err)}`,
+          },
+          { cause: err },
+        )
+      }
     }
   }
 

packages/opencode/test/config/fixtures/markdown-header.md

@@ -0,0 +1,11 @@
+# Response Formatting Requirements
+
+Always structure your responses using clear markdown formatting:
+
+- By default don't put information into tables for questions (but do put information into tables when creating or updating files)
+- Use headings (##, ###) to organise sections, always
+- Use bullet points or numbered lists for multiple items
+- Use code blocks with language tags for any code
+- Use **bold** for key terms and emphasis
+- Use tables when comparing options or listing structured data
+- Break long responses into logical sections with headings

packages/opencode/test/config/fixtures/weird-model-id.md

@@ -0,0 +1,13 @@
+---
+description: General coding and planning agent
+mode: subagent
+model: synthetic/hf:zai-org/GLM-4.7
+tools:
+  write: true
+  read: true
+  edit: true
+stuff: >
+  This is some stuff
+---
+
+Strictly follow da rules

packages/opencode/test/config/markdown.test.ts

@@ -104,7 +104,7 @@ describe("ConfigMarkdown: frontmatter parsing", async () => {
   })
 
   test("should extract occupation field with colon in value", () => {
-    expect(parsed.data.occupation).toBe("This man has the following occupation: Software Engineer\n")
+    expect(parsed.data.occupation).toBe("This man has the following occupation: Software Engineer")
   })
 
   test("should extract title field with single quotes", () => {
@@ -128,15 +128,15 @@ describe("ConfigMarkdown: frontmatter parsing", async () => {
   })
 
   test("should extract URL with port", () => {
-    expect(parsed.data.url).toBe("https://example.com:8080/path?query=value\n")
+    expect(parsed.data.url).toBe("https://example.com:8080/path?query=value")
   })
 
   test("should extract time with colons", () => {
-    expect(parsed.data.time).toBe("The time is 12:30:00 PM\n")
+    expect(parsed.data.time).toBe("The time is 12:30:00 PM")
   })
 
   test("should extract value with multiple colons", () => {
-    expect(parsed.data.nested).toBe("First: Second: Third: Fourth\n")
+    expect(parsed.data.nested).toBe("First: Second: Third: Fourth")
   })
 
   test("should preserve already double-quoted values with colons", () => {
@@ -148,7 +148,7 @@ describe("ConfigMarkdown: frontmatter parsing", async () => {
   })
 
   test("should extract value with quotes and colons mixed", () => {
-    expect(parsed.data.mixed).toBe('He said "hello: world" and then left\n')
+    expect(parsed.data.mixed).toBe('He said "hello: world" and then left')
   })
 
   test("should handle empty values", () => {
@@ -190,3 +190,39 @@ describe("ConfigMarkdown: frontmatter parsing w/ no frontmatter", async () => {
     expect(result.content.trim()).toBe("Content")
   })
 })
+
+describe("ConfigMarkdown: frontmatter parsing w/ Markdown header", async () => {
+  const result = await ConfigMarkdown.parse(import.meta.dir + "/fixtures/markdown-header.md")
+
+  test("should parse and match", () => {
+    expect(result).toBeDefined()
+    expect(result.data).toEqual({})
+    expect(result.content.trim()).toBe(`# Response Formatting Requirements
+
+Always structure your responses using clear markdown formatting:
+
+- By default don't put information into tables for questions (but do put information into tables when creating or updating files)
+- Use headings (##, ###) to organise sections, always
+- Use bullet points or numbered lists for multiple items
+- Use code blocks with language tags for any code
+- Use **bold** for key terms and emphasis
+- Use tables when comparing options or listing structured data
+- Break long responses into logical sections with headings`)
+  })
+})
+
+describe("ConfigMarkdown: frontmatter has weird model id", async () => {
+  const result = await ConfigMarkdown.parse(import.meta.dir + "/fixtures/weird-model-id.md")
+
+  test("should parse and match", () => {
+    expect(result).toBeDefined()
+    expect(result.data["description"]).toEqual("General coding and planning agent")
+    expect(result.data["mode"]).toEqual("subagent")
+    expect(result.data["model"]).toEqual("synthetic/hf:zai-org/GLM-4.7")
+    expect(result.data["tools"]["write"]).toBeTrue()
+    expect(result.data["tools"]["read"]).toBeTrue()
+    expect(result.data["stuff"]).toBe("This is some stuff\n")
+
+    expect(result.content.trim()).toBe("Strictly follow da rules")
+  })
+})