🧑‍💻 refactor: Clarify Code Artifact Guidance

[danny-avila]

Summary

I clarified code execution artifact guidance and restored a compact generated-file summary without listing every file.

• Added shared guidance for persistent /mnt/data artifacts, same-call /tmp scratch files, and conventional supported extensions.
• Appended compact session-file counts to code, bash, and programmatic tool outputs while keeping exact file refs in artifacts.
• Updated tests and stale session-file comments so the executor behavior is clear.

Change Type

• Bug fix (non-breaking change which fixes an issue)
• Documentation update

Testing

• npx jest src/tools/__tests__/ProgrammaticToolCalling.test.ts src/tools/__tests__/BashExecutor.test.ts --runInBand
• git diff --check

Test Configuration:

• Node dependencies borrowed from the main worktree via ignored node_modules symlink for the dedicated worktree test run.

Checklist

• My code adheres to this project's style guidelines
• I have performed a self-review of my own code
• I have commented in any complex areas of my code
• I have made pertinent documentation changes
• My changes do not introduce new warnings
• I have written tests demonstrating that my changes are effective or that my feature works
• Local unit tests pass with my changes

[danny-avila]

Post-open live testing against http://localhost:3112/v1:

• Verified raw /exec bash artifact behavior: supported /mnt/data/*.txt persisted, /tmp scratch did not persist across calls, and reinjection worked when response file refs were converted to CodeEnvFile shape.
• Verified .raw is not persisted while .json is persisted from the same /mnt/data run, matching CodeAPI's supported-extension allowlist.
• Verified /exec/programmatic Python and bash artifact behavior, including files generated after a tool-call continuation.
• Verified bash PTC continuation directly with lang: "bash" and a get_weather tool.
• Ran live @librechat/agents PTC integration against localhost. The full run passed the first 4 cases, then hit local CodeAPI 429 rate limiting; I reran the remaining 5 cases in spaced batches and they passed.
• Ran a temporary local Jest smoke for run_tools_with_bash against localhost and confirmed the new compact session-file summary appears without listing the generated filename. Removed the temporary test file after the smoke.

[danny-avila]

Added one terse bash-specific guidance sentence after the live LLM smoke exposed literal \n CSV writes:

For multi-line file writes, use heredoc or printf; echo "\\n" may stay literal.

Re-ran focused tests:

• npx jest src/tools/__tests__/ProgrammaticToolCalling.test.ts src/tools/__tests__/BashExecutor.test.ts --runInBand

[danny-avila]

@codex review

[danny-avila]

Reran the real LLM smoke after the terse bash multiline hint.

Prompt did not mention /mnt/data, /tmp, heredoc/printf, supported extensions, persistence, or Markdown image behavior. The model wrote the CSV with a heredoc on the first attempt:

cat <<EOF > /mnt/data/support_ticket_categories.csv
Category,Count
Login Issues,15
Payment Problems,9
Bug Reports,12
Feature Requests,7
Account Management,10
Other,5
EOF

It also saved the chart as /mnt/data/support_ticket_categories.png, reused the known CSV path on the follow-up without ls/find, avoided /tmp for persistent artifacts, and did not wrap the image in Markdown or invent a download link.

One unrelated bash-mode stumble remains: on the follow-up it first pasted Python statements directly into bash, got a syntax error, then self-corrected to python3 -c. The multiline file-write behavior improved as intended.

[chatgpt-codex-connector]

💡 Codex Review

Here are some automated review suggestions for this pull request.

Reviewed commit: 06efcdd46d

ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

• Open a pull request for review
• Mark a draft as ready
• Comment "@codex review".

If Codex has suggestions, it will comment; otherwise it will react with 👍.

Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".

[danny-avila]

Condensed the code-artifact/tool guidance while keeping the two live-smoke fixes:

• Shortened the shared artifact rule to one compact /mnt/data + extension + /tmp sentence.
• Shortened the PTC state/async rules and removed duplicated generated-file availability wording.
• Replaced the previous bash multiline note with one compact shell guard: Bash: multi-line files use heredoc/printf; run Python via python3 -c/heredoc, not bare Python.

Validation:

• npx jest src/tools/__tests__/ProgrammaticToolCalling.test.ts src/tools/__tests__/BashExecutor.test.ts --runInBand
• Real LLM smoke against http://localhost:3112/v1 with no explicit artifact/path/heredoc/Python-wrapper instructions in the prompt.

Live-smoke result: the model wrote the CSV with a heredoc, wrote the PNG under /mnt/data, reused the known CSV path in the follow-up, and used python3 -c instead of pasting bare Python into bash. It avoided /tmp, ls/find, Markdown image wrapping, and download-link wording.

[danny-avila]

@codex review

[chatgpt-codex-connector]

Codex Review: Didn't find any major issues. 🎉

ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

• Open a pull request for review
• Mark a draft as ready
• Comment "@codex review".

If Codex has suggestions, it will comment; otherwise it will react with 👍.

Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".

[danny-avila]

Added a small PTC inner-tool output contract after live testing against the remote CodeAPI-backed engine. Direct continuation checks confirmed Python receives object tool results as dicts and bash receives object/string tool results as JSON on stdout. Live LLM smoke before the change showed Python sometimes passed a positional dict and bash preserved JSON quotes for string results; after the added guidance, Python used keyword args on the first try and bash used jq -r . to unquote string results.

Validation:

• Remote CodeAPI /exec/programmatic direct continuation checks for Python and bash
• Live agent smoke using run_tools_with_code and run_tools_with_bash against http://localhost:3112/v1
• npx jest src/tools/__tests__/ProgrammaticToolCalling.test.ts src/tools/__tests__/BashExecutor.test.ts src/tools/__tests__/ToolNode.outputReferences.test.ts --runInBand
• npx eslint src/tools/ProgrammaticToolCalling.ts src/tools/BashProgrammaticToolCalling.ts src/tools/__tests__/ProgrammaticToolCalling.test.ts
• npx tsc --noEmit

[danny-avila]

@codex review

[chatgpt-codex-connector]

💡 Codex Review

Here are some automated review suggestions for this pull request.

Reviewed commit: 5f77b8cdb7

ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

• Open a pull request for review
• Mark a draft as ready
• Comment "@codex review".

If Codex has suggestions, it will comment; otherwise it will react with 👍.

Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".

[danny-avila]

@codex review

[chatgpt-codex-connector]

Codex Review: Didn't find any major issues. Nice work!

ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

• Open a pull request for review
• Mark a draft as ready
• Comment "@codex review".

If Codex has suggestions, it will comment; otherwise it will react with 👍.

Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".