anomalyco/opencode
1
0
Fork 0
mirror of https://github.com/anomalyco/opencode.git synced 2026-07-22 18:25:32 -04:00
Code Issues 5.9k Packages Projects Releases 19 Wiki Activity

System prompts force ASCII-only output, breaking support for non-English languages #8783

New Issue
Open
opened 2026-02-16 18:10:50 -05:00 by yindo · 2 comments
yindo commented 2026-02-16 18:10:50 -05:00
Owner
Copy Link
Copy Source

Originally created by @diegonix on GitHub (Feb 7, 2026).

Originally assigned to: @thdxr on GitHub.

Description

The system prompts contain an instruction that causes LLMs to avoid non-ASCII characters by default:

Default to ASCII when editing or creating files. Only introduce non-ASCII or other Unicode characters when there is a clear justification and the file already uses them.

This instruction appears in all model-specific prompt files:

  • codex_header.txt
  • anthropic.txt
  • beast.txt
  • gemini.txt
  • qwen.txt
  • trinity.txt
  • copilot-gpt-5.txt

While the intent is to prevent encoding issues in code files, LLMs interpret this broadly and apply it to all output, including conversational text. This effectively breaks support for any language that uses diacritics or non-Latin scripts.

Affected Languages

Language Missing Characters Example
Portuguese ç, ã, á, é, í, ó, ú, ô, ê "ação" → "acao"
Spanish ñ, á, é, í, ó, ú, ü "año" → "ano"
French é, è, ê, ë, à, â, ç, ô, î, ï, û, ù "français" → "francais"
German ä, ö, ü, ß "größe" → "grosse"
Polish ą, ć, ę, ł, ń, ó, ś, ź, ż "żółć" → "zolc"
Turkish ç, ğ, ı, ö, ş, ü "güneş" → "gunes"
Nordic (Swedish, Norwegian, Danish) å, ä, ö, æ, ø "blåbär" → "blabar"
Vietnamese ă, â, đ, ê, ô, ơ, ư + tones Severely broken
Russian, Ukrainian, Greek Cyrillic, Greek alphabets Completely unusable
Chinese, Japanese, Korean CJK characters Completely unusable
Arabic, Hebrew RTL scripts Completely unusable

Plugins

none

OpenCode version

1.1.53

Steps to reproduce

  1. Configure opencode with GPT-5.2

  2. Add a user prompt requesting output in a non-English language, e.g.:
    Always respond in Brazilian Portuguese with correct spelling and accents.

  3. Ask any question

  4. Observe that the response lacks proper diacritics/accents
    Expected: "A função executa a ação corretamente."
    Actual: "A funcao executa a acao corretamente."

Suggested Fix

Scope the ASCII constraint to code file editing only, and explicitly allow Unicode in conversational output:

## Editing constraints
-- Default to ASCII when editing or creating files. Only introduce non-ASCII or other Unicode characters when there is a clear justification and the file already uses them.
++ When editing or creating source code files, prefer ASCII for identifiers and comments unless the project already uses non-ASCII characters. This constraint does NOT apply to conversational responses, documentation, or user-facing text—use proper Unicode characters as appropriate for the language.



### Screenshot and/or share link

_No response_

### Operating System

Ubuntu 24.04.4

### Terminal

Ghostty 1.2.3
Originally created by @diegonix on GitHub (Feb 7, 2026). Originally assigned to: @thdxr on GitHub. ### Description The system prompts contain an instruction that causes LLMs to avoid non-ASCII characters by default: `Default to ASCII when editing or creating files. Only introduce non-ASCII or other Unicode characters when there is a clear justification and the file already uses them.` This instruction appears in **all** model-specific prompt files: - `codex_header.txt` - `anthropic.txt` - `beast.txt` - `gemini.txt` - `qwen.txt` - `trinity.txt` - `copilot-gpt-5.txt` While the intent is to prevent encoding issues in code files, LLMs interpret this broadly and apply it to **all output**, including conversational text. This effectively breaks support for any language that uses diacritics or non-Latin scripts. ## Affected Languages | Language | Missing Characters | Example | |----------|-------------------|---------| | Portuguese | ç, ã, á, é, í, ó, ú, ô, ê | "ação" → "acao" | | Spanish | ñ, á, é, í, ó, ú, ü | "año" → "ano" | | French | é, è, ê, ë, à, â, ç, ô, î, ï, û, ù | "français" → "francais" | | German | ä, ö, ü, ß | "größe" → "grosse" | | Polish | ą, ć, ę, ł, ń, ó, ś, ź, ż | "żółć" → "zolc" | | Turkish | ç, ğ, ı, ö, ş, ü | "güneş" → "gunes" | | Nordic (Swedish, Norwegian, Danish) | å, ä, ö, æ, ø | "blåbär" → "blabar" | | Vietnamese | ă, â, đ, ê, ô, ơ, ư + tones | Severely broken | | Russian, Ukrainian, Greek | Cyrillic, Greek alphabets | Completely unusable | | Chinese, Japanese, Korean | CJK characters | Completely unusable | | Arabic, Hebrew | RTL scripts | Completely unusable | ### Plugins none ### OpenCode version 1.1.53 ### Steps to reproduce 1. Configure opencode with GPT-5.2 2. Add a user prompt requesting output in a non-English language, e.g.: `Always respond in Brazilian Portuguese with correct spelling and accents.` 3. Ask any question 5. Observe that the response lacks proper diacritics/accents **Expected:** "A função executa a ação corretamente." **Actual:** "A funcao executa a acao corretamente." ## Suggested Fix Scope the ASCII constraint to **code file editing only**, and explicitly allow Unicode in conversational output: ```diff ## Editing constraints -- Default to ASCII when editing or creating files. Only introduce non-ASCII or other Unicode characters when there is a clear justification and the file already uses them. ++ When editing or creating source code files, prefer ASCII for identifiers and comments unless the project already uses non-ASCII characters. This constraint does NOT apply to conversational responses, documentation, or user-facing text—use proper Unicode characters as appropriate for the language. ### Screenshot and/or share link _No response_ ### Operating System Ubuntu 24.04.4 ### Terminal Ghostty 1.2.3
yindo added the bug label 2026-02-16 18:10:50 -05:00
yindo commented 2026-02-16 18:10:50 -05:00
Author
Owner
Copy Link
Copy Source

@github-actions[bot] commented on GitHub (Feb 7, 2026):

This issue might be a duplicate of existing issues. Please check:

  • #12441: Project avatar fallback doesn't support emoji characters - While focused on avatar rendering, it reveals the same underlying issue: JavaScript string handling and Unicode character support (emoji handling with surrogate pairs)
  • #12197: garbled characters in darwin vscode terminal - Related to character rendering issues, though in a different context

The core issue in #12609 is that system prompts impose ASCII-only constraints globally, affecting all output including conversational text in non-English languages. This is distinct from the avatar emoji rendering issue but shares related Unicode handling concerns.

Feel free to ignore if none of these address your specific case.

@github-actions[bot] commented on GitHub (Feb 7, 2026): This issue might be a duplicate of existing issues. Please check: - #12441: Project avatar fallback doesn't support emoji characters - While focused on avatar rendering, it reveals the same underlying issue: JavaScript string handling and Unicode character support (emoji handling with surrogate pairs) - #12197: garbled characters in darwin vscode terminal - Related to character rendering issues, though in a different context The core issue in #12609 is that system prompts impose ASCII-only constraints globally, affecting all output including conversational text in non-English languages. This is distinct from the avatar emoji rendering issue but shares related Unicode handling concerns. Feel free to ignore if none of these address your specific case.
yindo commented 2026-02-16 18:10:50 -05:00
Author
Owner
Copy Link
Copy Source

@diegonix commented on GitHub (Feb 16, 2026):

I know there are more important features and bugs to address, but OpenCode is now a global project with users all around the world. Please fix this for us.
There is a damn system_prompt preventing the LLM from doing what it does best - writing correctly in a human language.

Image
@diegonix commented on GitHub (Feb 16, 2026): I know there are more important features and bugs to address, but OpenCode is now a global project with users all around the world. Please fix this for us. There is a damn system_prompt preventing the LLM from doing what it does best - writing correctly in a human language. <img width="1001" height="95" alt="Image" src="https://github.com/user-attachments/assets/dbfb18b2-20ca-4ecd-9b02-29f969f36590" />
Sign in to join this conversation.
No Branch/Tag Specified
Branches Tags
Labels
Clear labels
acp
bad
bug
discussion
docs
documentation
enhancement
extension
good first issue
growth
help-wanted
mcp
model-problem
needs:compliance
nix
opentui
perf
pull-request
Mirrored from GitHub Pull Request
 tui
web
windows
zen
No Label bug
Milestone
No items
No Milestone
Projects
Clear projects
No project
Assignees
Clear assignees
yindo
No Assignees
1 Participants
Notifications
Due Date
No due date set.
Dependencies

No dependencies set.

Reference: anomalyco/opencode#8783
Delete Branch "%!s()"

Deleting a branch is permanent. Although the deleted branch may continue to exist for a short time before it actually gets removed, it CANNOT be undone in most cases. Continue?