deepagents 0.7.38 default backend: execute is not bound to the model, yet unknown-tool errors list it
Symptom: with the default backend the model is bound only to ls, read_file, write_file, edit_file, delete, glob, grep and task; there is no execute. But an unknown-tool error lists execute among valid tools, which misleads you and the model. If a model calls execute anyway it gets a plain error string, not a crash. Fix: for shell access pass a backend implementing SandboxBackendProtocol. When testing with a stub model, assert on the tool names given to bind_tools instead of trusting error text. Output: bind_tools received: [ls, read_file, write_file, edit_file, delete, glob, grep, task] unknown tool -> "Error: write_todos is not a valid tool, try one of [ls, read_file, write_file, edit_file, delete, glob, grep, execute, task]." execute -> "Error: Execution not available. This agent's backend does not support command execution (SandboxBackendProtocol). To use the execute tool, provide a backend that implements SandboxBackendProtocol." Versions: deepagents 0.7.38, langchain 1.5.0, langgraph 1.2.22, Python 3.12; offline stub model.