fix(moa): add explicit warnings to reference prompt against claiming tool execution

[liuhao1024]

What does this PR do?

Prevents MoA (Mixture of Agents) reference models from fabricating tool execution in their text output. Reference models are advisory-only and should never claim to have executed commands, downloaded files, or accessed URLs — even as hypothetical statements.

Despite the existing _REFERENCE_SYSTEM_PROMPT stating "you cannot call tools", models still hallucinate execution in their responses (e.g., "I ran curl and got 404"). This fix adds:

1. A prominent CRITICAL warning against claiming/implying execution
2. Concrete bad/good examples showing what to avoid
3. A trailing reminder "NEVER claim to have executed anything"

This addresses the aggregator's inability to distinguish real tool outputs from fabricated claims, which erodes trust in the entire MoA pipeline.

Related Issue

Fixes #61452

Type of Change

• 🐛 Bug fix (non-breaking change that fixes an issue)

Changes Made

• agent/moa_loop.py: Enhanced _REFERENCE_SYSTEM_PROMPT with explicit warnings and examples against claiming tool execution (12 lines added)
• tests/agent/test_moa_reference_system_prompt.py: Added regression tests verifying the prompt contains necessary constraints (2 tests)

How to Test

Run the new regression tests to verify the prompt contains the required warnings:

pytest tests/agent/test_moa_reference_system_prompt.py -v

Observed result: both tests pass, confirming:

• The prompt explicitly states reference models cannot execute anything
• The prompt warns against claiming/implying execution
• The prompt provides bad/good examples

End-to-end: When using MoA mode, reference model outputs should no longer contain claims like "I ran curl" or "I downloaded the file successfully."

Checklist

Code

• I've read the Contributing Guide
• My commit messages follow Conventional Commits (fix(scope):, feat(scope):, etc.)
• I searched for existing PRs to make sure this isn't a duplicate
• My PR contains only changes related to this fix/feature (no unrelated commits)
• I've run pytest tests/ -q and all tests pass
• I've added tests for my changes (required for bug fixes, strongly encouraged for features)
• I've tested on my platform: macOS 15.2

Documentation & Housekeeping

• I've updated relevant documentation (README, docs/, docstrings) — or N/A
• I've updated cli-config.yaml.example if I added/changed config keys — or N/A
• I've updated CONTRIBUTING.md or AGENTS.md if I changed architecture or workflows — or N/A
• I've considered cross-platform impact (Windows, macOS) per the compatibility guide — or N/A
• I've updated tool descriptions/schemas if I changed tool behavior — or N/A

[tonydwb]

Code Review Summary

Verdict: LGTM

What the PR Does

Add explicit warnings to the MOA reference prompt against claiming tool execution.

Assessment

• Small, focused doc/instruction fix.

---

Reviewed by Hermes Agent