fix: mutable default arg in Memory and improve robustness

[KeloYuan]

Summary

This PR fixes a critical bug and adds several robustness improvements:

🐛 Critical Bug Fix

• schema.py: Memory.messages used a mutable default argument (messages: List[Message] = []). In Python, mutable defaults are shared across all instances of a class — meaning multiple Memory() instances would share the same message list, causing cross-contamination between agents. Fixed by using Field(default_factory=list).

🔧 Robustness Improvements

• agent.py: Add max_steps validation — MiniAgent(max_steps=0) or negative values now raise ValueError immediately instead of silently producing no output.
• agent.py: Handle empty tool output gracefully — result_content[:100] on an empty string produced "" with trailing ..., now shows (empty) instead.

🔒 Security

• tools.py: Add warnings.warn() when BashExecutor executes commands with shell=True, so users are aware of the security implications.

📦 Public API

• __init__.py: Export Role and LLMResponse so users can import the full public API from mini_agent without reaching into submodules.

[Copilot]

Pull request overview

This PR addresses a shared-state bug in the agent memory model and adds a few small safety/robustness improvements across the agent runtime and public API surface.

Changes:

• Fix Memory.messages mutable default by switching to Field(default_factory=list).
• Add max_steps validation and improve tool-output previewing for empty outputs.
• Warn on BashExecutor usage of shell=True and export additional public API symbols (Role, LLMResponse) from mini_agent.

Reviewed changes

Copilot reviewed 4 out of 4 changed files in this pull request and generated no comments.

File	Description
mini_agent/tools.py	Adds a security warning when executing shell commands via shell=True.
mini_agent/schema.py	Fixes shared mutable default for Memory.messages using default_factory.
mini_agent/agent.py	Validates max_steps and improves logging output for empty tool results.
mini_agent/init.py	Exposes Role and LLMResponse in the package public API.

Comments suppressed due to low confidence (1)

mini_agent/tools.py:156

• BashExecutor.execute is declared async but uses blocking subprocess.run(...), which will block the event loop and can stall concurrent agent steps. Consider running the subprocess via asyncio.create_subprocess_shell/exec or offloading the blocking call with asyncio.to_thread to keep the async API non-blocking.

            result = subprocess.run(
                command,
                shell=True,
                capture_output=True,
                text=True,
                timeout=30
            )

---

💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.

[chatgpt-codex-connector]

💡 Codex Review

Here are some automated review suggestions for this pull request.

Reviewed commit: a3504e58cb

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".

[chatgpt-codex-connector]

Avoid failing BashExecutor under warnings-as-errors

Emitting warnings.warn(...) inside BashExecutor.execute can make every bash call fail in environments that run with PYTHONWARNINGS=error or -W error (common in CI/tests): warn raises UserWarning, it is caught by the broad except Exception, and the method returns success=False before/without running the command. This turns a diagnostic warning into a functional regression for those runtimes.

Useful? React with 👍 / 👎.