Hi,

This PR adds the small docs-only troubleshooting note we discussed in #14.

Summary

• Adds docs/multi_tool_agent_failure_modes.md, a short troubleshooting page focused on multi-tool agent failure modes when running ToolBench, API-Bank, ToolHop, GAIA, HLE and other benchmarks.
• Links the new page from the README under a short “Troubleshooting multi tool failures” section near the evaluation flags.

The note is intentionally compact and DeepAgent-specific:

• Starts with a one-screen quick checklist to run when the agent looks stuck or keeps calling strange tools.
• Breaks down typical failure patterns (wrong tool, argument mismatch, environment/config mismatch, tool outputs ignored, etc.) into “symptom → likely cause → minimal checks”.
• Keeps all examples aligned with the existing configuration flags and scripts in this repo.

Scope

• Docs only, no code or config changes.
• Does not affect any training or evaluation scripts.

Testing

• Rendered both the README and docs/multi_tool_agent_failure_modes.md in GitHub’s preview to check formatting and links.

Closes #14.

Thanks for reviewing!