[FEATURE] Add all tool details under gen_ai.agent.tools attribute

[kiranramnath007]

Problem Statement

Requesting Strands team to emit the following tool details in the invoke_agent.metadata. attributes.gen_ai.agent.tools.

Currently, this information is only available in the tool_invocation span for the single tool that is invoked, but we need some extra details for AgentCore Evaluation to support the following evaluation needs:

• For ToolSelectionAccuracy evaluator, we need to provide the list of all available tools to help the LLM-as-a-Judge decide if the correct tool was chosen.

• For ToolParamaterAccuracy evaluator, we need to provide the list of all required parameters (and not just the ones sent in the request) to help the LLMJ decide if the parameters were inputted correctly

Proposed Solution

Add the following details under invoke_agent.metadata.attributes.gen_ai.agent.tools

List[ToolConfig]

where each ToolConfig consists of

• Tool Name
• Tool Input parameters + types
• Tool output parameter + type
• Tool description / docstring

Use Case

This information is required to input into the LLM-as-a-Judge prompt for AgentCoreEvaluation, and also Strands Evaluation feature.

Alternatives Solutions

No response

Additional Context

No response

[poshinchen]

Acceptance Criteria:

1. Do other frameworks have the same structure to attach all json_schemas of the available tools?
   1.1 If so, we can follow-up on adding the spans
   1.2 If not, we should add a flag for users to opt-in the spans as it will increase the amount of spans / data size to the user.