fix: add include_default_tools parameter to LLMAgent to prevent unwanted inbuilt tool calls

[saakshisinghal14]

Fixes #254

Summary

When an agent registers only domain-specific tools (e.g. adopt_opinion), the LLM executor can still call inbuilt tools like move_one_step — even when the simulation has nothing to do with movement.

This happens because ToolManager auto-registers inbuilt tools globally, with no way to opt out. When the executor must choose a tool, it may select these unintended defaults.

Changes

• Added INBUILT_TOOL_NAMES (frozenset) in inbuilt_tools.py to centrally define default tool names
• Added include_default_tools: bool = True parameter to LLMAgent.__init__()
  • True (default): preserves existing behavior (fully backward compatible)
  • False: filters out inbuilt tools (move_one_step, teleport_to_location, speak_to) from the agent’s ToolManager, leaving only user-registered tools
• Added tests covering:
  • default behavior (inbuilt tools present)
  • opt-out behavior (inbuilt tools removed)
  • preservation of user-defined tools

Implementation Note

Filtering is applied after tool registration, ensuring compatibility with the existing tool registration flow while allowing selective removal of default tools per agent.

Usage

agent = LLMAgent(
    model=model,
    reasoning=CoTReasoning,
    include_default_tools=False,  # only domain tools visible to LLM
)

[coderabbitai]

Important

Review skipped

Auto reviews are disabled on this repository. Please check the settings in the CodeRabbit UI or the .coderabbit.yaml file in this repository. To trigger a single review, invoke the @coderabbitai review command.

⚙️ Run configuration

Configuration used: Organization UI

Review profile: CHILL

Plan: Pro

Run ID: b024c973-b477-41f1-a0ba-e699421f3774

You can disable this status message by setting the reviews.review_status to false in the CodeRabbit configuration file.

Use the checkbox below for a quick retry:

• 🔍 Trigger review

✨ Finishing Touches

🧪 Generate unit tests (beta)

• Create PR with unit tests

---

Thanks for using CodeRabbit! It's free for OSS, and your support helps us grow. If you like it, consider giving us a shout-out.

❤️ Share

• X
• Mastodon
• Reddit
• LinkedIn

_{Comment @coderabbitai help to get the list of available commands and usage tips.}

[BhoomiAgrawal12]

Hey @saakshisinghal14, PR #167 by @yashhzd addresses the same issue in the same way. It also adds an opt-out flag to LLMAgent.init and passes it through to ToolManager.

I would suggest since PR #167 was first and has similar intent, it might be worth either closing this one in its favor, or leaving a comment on #167 discussing the differences so the maintainers can make an informed call. Thanks for putting in the work on this!

[saakshisinghal14]

Thanks for pointing this out @BhoomiAgrawal12 . I wasn't aware of PR #167 when I opened this — it predates my involvement with the repo.

Looking at it, #167 takes a more thorough approach by adding the flag to ToolManager itself and including a remove_tool() method, which is cleaner than my post-creation filtering. I'm happy to defer to #167 if the maintainers prefer that approach.