Using temp files for code instead of putting code on command line eliminates permission prompts

[vsekhar]

The skill tells the agent to provide code on the commandline.

bash scripts/execute-code.sh -c "code"
bash scripts/execute-code.sh <<'EOF'

Even after giving Claude Code permission to run execute-code.sh, this usage triggers permission prompts for any complex code block.

The fix is to tell the agent to write its code to a temporary file, execute it, and then clean up the temporary file.

Edit(edit-cell-s2j5.tmp.py)
bash scripts/execute-code.sh edit-cell-s2j5.tmp.py; rm edit-cell-s2j5.tmp.py

Writing the temporary file to the current directory eliminates the permission prompt for access to /tmp or similar. Adding the .tmp.py extension makes it easier to cleanup if the agent misses doing so and allows these files to be added to .gitignore. Permission prompts are suppressed for code of any complexity (once execute-code.sh is allowed).

[manzt]

Hi there, thanks for the suggestion. The goal of execute-code is to provide a similar API surface to the Python CLI, which agents are good at using. The skill now includes (#51):

When code already lives in a file, pass the file path:

bash scripts/execute-code.sh --url http://localhost:2718 /tmp/code.py

which hopefully highlights this possibility in the SKILL itself, instead of requiring the agent to dig in the CLI.