Add a dedicated read_file tool for Codex (apply_patch is not enough)

What feature would you like to see?

Hello team,

Currently, Codex only provides apply_patch for writing and editing files, but there is no dedicated and explicit tool for reading file contents.

This limitation causes many repetitive issues in real-world workflows:

• The biggest issue is that Codex spends too much time reasoning about how to read files and how to execute file-reading commands
• Because there is no suitable read_file tool, Codex is forced to rely on indirect mechanisms or guesswork, leading to unnecessary overhead
• Encoding errors frequently occur when files contain Vietnamese or other Unicode characters (UTF-8)
• Output can become subtly incorrect (broken characters, missing lines, wrong context)
• A lot of developer time is wasted debugging issues caused by unreliable file reading

I would like to propose a dedicated read_file (or equivalent) tool with the following characteristics:

• Explicitly reads file contents without requiring inference
• Preserves data in a lossless manner
• Fully supports UTF-8 / Unicode (including Vietnamese) (I know there is a UTF-8 shell in /experimental, but it still has character issues)
• Can read the entire file or specific line ranges
• Is clearly separated from write/edit tools like apply_patch

Additional information

In practice, directly reading files is both faster and safer than reconstructing file state through patch-based workflows.

At the moment, Codex often spends more time reasoning about how to perform file-reading operations than actually executing meaningful tasks. This negatively impacts iteration speed, accuracy, and overall developer productivity.

A proper read_file tool would:

• Eliminate unnecessary reasoning overhead related to file access
• Improve correctness and reproducibility
• Reduce encoding-related errors
• Save tokens and developer time
• Make Codex workflows much closer to real-world IDE usage

This would be a major quality-of-life improvement for anyone using Codex on real codebases, especially those containing non-English or Unicode-heavy content.