From 9da72041495785edee05d9f9f3860f80baf61e76 Mon Sep 17 00:00:00 2001
From: EQCommunication <97274751+EQCommunication@users.noreply.github.com>
Date: Wed, 18 Feb 2026 00:30:59 +0900
Subject: [PATCH] fix: sanitize lone surrogates in API request body to prevent
 JSON parse errors

Anthropic API's serde_json strictly enforces RFC 8259 and rejects JSON containing lone surrogate characters. JavaScript's JSON.stringify() allows lone surrogates (per ECMA-262), so when tool outputs contain non-UTF-8 data (binary files, terminal output), the API returns a 400 error.

This adds sanitization in the custom fetch wrapper to replace lone surrogates (U+D800-U+DFFF) with U+FFFD before sending the request body. Uses String.prototype.toWellFormed() (ES2024) with regex fallback.

Fixes the 'no low surrogate in string' JSON parse error reported across multiple tools. Related: anthropics/claude-code#1709

Ultraworked with [Sisyphus](https://github.com/code-yeongyu/oh-my-opencode)

Co-authored-by: Sisyphus <clio-agent@sisyphuslabs.ai>
---
 packages/opencode/src/provider/provider.ts    |  5 +++
 .../src/util/sanitize-surrogates.test.ts      | 38 +++++++++++++++++++
 .../opencode/src/util/sanitize-surrogates.ts  |  6 +++
 3 files changed, 49 insertions(+)
 create mode 100644 packages/opencode/src/util/sanitize-surrogates.test.ts
 create mode 100644 packages/opencode/src/util/sanitize-surrogates.ts

diff --git a/packages/opencode/src/provider/provider.ts b/packages/opencode/src/provider/provider.ts
index cdbad663784..30ca52f1016 100644
--- a/packages/opencode/src/provider/provider.ts
+++ b/packages/opencode/src/provider/provider.ts
@@ -5,6 +5,7 @@ import { Config } from "../config/config"
 import { mapValues, mergeDeep, omit, pickBy, sortBy } from "remeda"
 import { NoSuchModelError, type Provider as SDK } from "ai"
 import { Log } from "../util/log"
+import { sanitizeSurrogates } from "../util/sanitize-surrogates"
 import { BunProc } from "../bun"
 import { Plugin } from "../plugin"
 import { ModelsDev } from "./models"
@@ -1093,6 +1094,10 @@ export namespace Provider {
           }
         }
 
+        if (typeof opts.body === "string") {
+          opts.body = sanitizeSurrogates(opts.body)
+        }
+
         return fetchFn(input, {
           ...opts,
           // @ts-ignore see here: https://github.com/oven-sh/bun/issues/16682
diff --git a/packages/opencode/src/util/sanitize-surrogates.test.ts b/packages/opencode/src/util/sanitize-surrogates.test.ts
new file mode 100644
index 00000000000..d18b8fb8f54
--- /dev/null
+++ b/packages/opencode/src/util/sanitize-surrogates.test.ts
@@ -0,0 +1,38 @@
+import { describe, test, expect } from "bun:test"
+import { sanitizeSurrogates } from "./sanitize-surrogates"
+
+describe("sanitizeSurrogates", () => {
+  test("replaces lone high surrogate", () => {
+    expect(sanitizeSurrogates("\uD800")).toBe("\uFFFD")
+  })
+
+  test("replaces lone low surrogate", () => {
+    expect(sanitizeSurrogates("\uDC00")).toBe("\uFFFD")
+  })
+
+  test("preserves valid surrogate pair", () => {
+    const emoji = "\uD83D\uDE00"
+    expect(sanitizeSurrogates(emoji)).toBe(emoji)
+  })
+
+  test("preserves normal text", () => {
+    expect(sanitizeSurrogates("hello world")).toBe("hello world")
+  })
+
+  test("preserves Korean text", () => {
+    expect(sanitizeSurrogates("안녕하세요")).toBe("안녕하세요")
+  })
+
+  test("preserves empty string", () => {
+    expect(sanitizeSurrogates("")).toBe("")
+  })
+
+  test("replaces surrogate in middle", () => {
+    expect(sanitizeSurrogates("hello\uD800world")).toBe("hello\uFFFDworld")
+  })
+
+  test("result is well-formed", () => {
+    const result = sanitizeSurrogates("test\uD800\uDBFF\uDC00data\uDFFF")
+    expect(result.isWellFormed()).toBe(true)
+  })
+})
diff --git a/packages/opencode/src/util/sanitize-surrogates.ts b/packages/opencode/src/util/sanitize-surrogates.ts
new file mode 100644
index 00000000000..1de3fdbbcd3
--- /dev/null
+++ b/packages/opencode/src/util/sanitize-surrogates.ts
@@ -0,0 +1,6 @@
+export function sanitizeSurrogates(s: string): string {
+  if (typeof s !== "string" || s.length === 0) return s
+  if (typeof s.isWellFormed === "function" && s.isWellFormed()) return s
+  if (typeof s.toWellFormed === "function") return s.toWellFormed()
+  return s.replace(/[\uD800-\uDBFF](?![\uDC00-\uDFFF])|(?<![\uD800-\uDBFF])[\uDC00-\uDFFF]/g, "\uFFFD")
+}