REPL mode file detection missing common pandas export methods (to_excel, to_json, etc.)

[AI-Flow-Logic]

Hi,

First of all, thank you for this great project!

Observation

I noticed that when Python REPL mode is enabled, files generated with to_excel() (and potentially other pandas export methods) don't seem to be captured
in the API response. The files array comes back empty even though the file is successfully created in /mnt/data/.

Interestingly, the same workflow with to_csv() works perfectly.

How to Reproduce

1. REPL mode enabled (REPL_ENABLED=true)
2. Execute via /exec:

   {"lang":"py","code":"import pandas as pd\ndf = pd.DataFrame({\"City\": [\"Tokyo\"], \"Pop\": [37400000]})\ndf.to_excel(\"/mnt/data/test.xlsx\",

index=False)\nprint("done")"}
3. Response: "files":[] — file not captured
4. Same test with to_csv works as expected

What I Found

Looking at src/services/execution/runner.py (around line 224), it seems like the REPL mode uses a keyword list to decide whether to scan for generated
files:

for kw in ["open(", "savefig", "to_csv", "write(", ".save("]

It looks like to_excel and a few other common pandas export methods might be missing from this list.

Possible Fix

Adding the missing keywords seems to resolve the issue on our side:

for kw in ["open(", "savefig", "to_csv", "to_excel", "to_json", "to_parquet", "to_html", "to_xml", "to_feather", "to_pickle", "write(", ".save(", "dump("]

dump( would also cover json.dump(), pickle.dump(), yaml.dump() patterns.

Of course, you may have a better approach in mind — just sharing what worked for us. Happy to submit a PR if that would be helpful.

Environment

• LibreCodeInterpreter: latest (GHCR)
• Python execution image: latest (GHCR)
• REPL_ENABLED=true
• STATE_PERSISTENCE_ENABLED=true