Add reporting for tools used

[cmungall]

Currently if you want to get a list of tools used it's necessary to parse structured_output - and this is not harmonized across coders

e.g. claude:

        - type: assistant
      message:
        id: msg_01Jj8Sq4nPSGGv4q8Etq5D8a
        type: message
        role: assistant
        model: claude-sonnet-4-20250514
        content:
        - type: tool_use
          id: toolu_019mJqdgpJSP1Z6UcfsMhx7s
          name: mcp__pubmed__get_paper_fulltext
          input:
            pmid: '35743164'
        stop_reason: null
        stop_sequence: null
        usage:
          input_tokens: 4
          cache_creation_input_tokens: 15333
          cache_read_input_tokens: 0
          output_tokens: 99
          service_tier: standard
      parent_tool_use_id: null
      session_id: 9aceb105-3b43-48f1-a67f-303a125c97fa
    - type: user
      message:
        role: user
        content:
        - type: tool_result
          content: MCP tool "get_paper_fulltext" response (37865 tokens) exceeds maximum
            allowed tokens (25000). Please use pagination, filtering, or limit parameters
            to reduce the response size.
          is_error: true
          tool_use_id: toolu_019mJqdgpJSP1Z6UcfsMhx7s

goose:

    - id: null
      role: assistant
      created: 1754353029
      content:
      - type: text
        text: I'll help you find information about diseases associated with ITPR1
          mutations from the specified PMID. Let me first retrieve the full text of
          that paper.
      - type: toolRequest
        id: toolu_01RbESTBH9tyWu9Q9uAVRjja
        toolCall:
          status: success
          value:
            name: pubmed__get_paper_fulltext
            arguments:
              pmid: '35743164'
    - id: null
      role: user
      created: 1754353029
      content:
      - type: toolResponse
        id: toolu_01RbESTBH9tyWu9Q9uAVRjja
        toolResult:
          status: success
          value:
          - type: text
            text: '<?xml version="1.0"  ?><!DOCTYPE pmc-articleset PUBLIC "-//NLM//DTD
            ...

[dragon-ai-agent]

Status Update on Tool Usage Reporting

I examined the current tool usage reporting implementation:

Current Status:

• ✅ Both Claude and Goose coders already implement tool usage parsing
• ✅ Harmonized ToolUse model exists in base_coder.py with standardized fields:
  • name: Tool name (e.g., "mcp__pubmed__get_paper_fulltext")
  • arguments: Tool input parameters
  • success: Whether tool call succeeded
  • error: Error message if failed
  • result: Tool execution result
• ✅ Coders populate tool_uses field in CoderOutput

Implementation Analysis:
Both coders parse their native structured output and convert to the common ToolUse format:

• Claude: Parses tool_use and tool_result from stream-json output
• Goose: Parses toolRequest and toolResponse from structured messages

Missing Coverage:
Need to verify tool usage parsing for:

• Gemini CLI
• Codex
• Qwen-coder
• Other coders in the registry

Access Pattern:
Users can now access harmonized tool usage via:

output = coder.run("query")
for tool_use in output.tool_uses or []:
    print(f"Tool: {tool_use.name}")
    print(f"Success: {tool_use.success}")
    if tool_use.error:
        print(f"Error: {tool_use.error}")

Status: Already implemented for main coders! Need to verify/add support for remaining coders.