NullCry 更新:修正長檔案截斷及 AI 檔案解析問題
NullCry Update: Fixing Long-File Truncation & AI File Parsing
NullCry CEO Omar Zahouri 表示,將於未來 48 小時內推出重大更新,修正 AI 生成專案中的長檔案截斷及檔案解析可靠性問題。更新將改善 CONT_PROMPT 的續寫、合併 CONTINUED sections,並以 “Surgical Stitch” 重建完整檔案。sanitizeOutput 亦將改良,避免程式碼中的有效反引號被誤判為程式碼區塊結尾。
I’m Omar Zahouri, CEO of NullCry.
In the next 48 hours, I’m releasing a major NullCry update focused on fixing some of the most important reliability issues in AI-generated projects.
The update includes:
- Long-file truncation fix Long files will now be accumulated through multiple generation passes without losing previously generated lines.
• Support for:
file: path CONTINUED
• The longest valid version of each file is preserved.
• A final “Surgical Stitch” reconstructs the complete file from accumulated drafts.
- Fixing “The AI returned no valid files”
NullCry will now recognize more common AI output formats, including:
file: index.html
file index.html
index.html
html / javascript / python code blocks
It will also detect filename/path hints such as:
styles.css
If files still cannot be extracted from a response, NullCry can automatically request the AI to reformat the response into explicit file blocks.
- More reliable file continuation
CONT_PROMPT is being improved to continue generation precisely from the interruption point.
CONTINUED sections will also be merged automatically instead of being treated as separate incomplete files.
- Output sanitization improvements
sanitizeOutput is being improved so legitimate backticks inside source code are not accidentally interpreted as the end of a generated code block.
The goal is simple:
NullCry should not just generate code.
It should reliably understand the requested project, work across files, handle large files, continue interrupted generations, and reconstruct the final project without silently losing code.
This is the next step in making NullCry more reliable for real software development.
— Omar Zahouri
CEO of NullCry
來源:DEV Community · dev.to