[BUG] Windows: text files in system code page encoding (GBK/Shift-JIS) produce garbled output when read as UTF-8 #31765

Description

On non-English Windows systems, text files (including AGENTS.md) may be saved in the system's active code page encoding (e.g., GBK for Chinese, Shift-JIS for Japanese) instead of UTF-8. The fs.readFileString() function always decodes as UTF-8, producing garbled characters (mojibake) for non-UTF-8 files.

Affected code

packages/core/src/fs-util.ts - the readFileString method delegates to the platform's fs.readFileString() which assumes UTF-8.

Also packages/opencode/src/session/instruction.ts:91 - the read helper used to load AGENTS.md uses fs.readFileString(filepath).

Impact

• Users who create AGENTS.md using Windows Notepad (which defaults to ANSI/GBK encoding on Chinese Windows) will see garbled text
• CI/test files saved in non-UTF-8 encodings will produce corrupted output
• This affects all file reading, not just instruction files

Suggested fix

Read the file as raw bytes, attempt UTF-8 decoding first, and if replacement characters (U+FFFD) are present, fall back to the system's active code page encoding detected via chcp.com.

Steps to reproduce

1. On a Chinese Windows system (code page 936)
2. Create a file with Chinese text using Notepad (saves as GBK by default)
3. fs.readFileString() decodes GBK bytes as UTF-8, producing garbled text like ���IJ���

OpenCode version

1.17.3

Operating System

Windows 11 (zh-CN)