Send your first trace.

Add TwoTail tracing to this agent. TwoTail accepts standard OpenTelemetry spans - no custom conventions required.

## Configuration
- Endpoint: https://www.twotail.ai/api/v1/traces
- API Key: <TWOTAIL_API_KEY> (created in the TwoTail app under API Keys)
- Auth Header: X-API-Key

## Trace Structure (important)
Generate ONE trace_id per agent run and use the SAME trace_id for ALL spans in that run. Each span gets its own unique span_id. Link child spans to parents via parentSpanId. This groups everything into a single trace so you can see the full execution flow.
- trace_id: 32 lowercase hex chars (16 bytes) - e.g., "4bf92f3577b34da6a3ce929d0e0e4736"
- span_id: 16 lowercase hex chars (8 bytes) - e.g., "00f067aa0ba902b7"
- parentSpanId: same format as span_id, or null/omit for root spans

## Sessions (multi-turn conversations)
If your agent runs multi-turn conversations, set gen_ai.conversation.id on the spans of each turn to the conversation/thread id. Use one trace_id per turn (run) and the SAME gen_ai.conversation.id across the whole conversation. TwoTail groups all traces sharing a gen_ai.conversation.id into one session.

## Standard OTel Attributes (recommended - the more you send, the more TwoTail can analyze)

For LLM calls:
- gen_ai.system: "openai" | "anthropic" | etc (auto-detects span as LLM type)
- gen_ai.request.model: model name (e.g., "gpt-4", "claude-3-sonnet")
- gen_ai.prompt: the input prompt/messages (important for debugging and analysis)
- gen_ai.completion: the output completion (important for debugging and analysis)
- gen_ai.usage.input_tokens: input token count
- gen_ai.usage.output_tokens: output token count

For tool calls:
- tool.name: name of the tool being called (auto-detects span as tool type)
- tool.parameters: input parameters (JSON)
- tool.result: output result

For evaluations (as child spans of the span being evaluated):
- Start the span name with "eval." or "eval_" (e.g., "eval.relevance", "eval_toxicity") - auto-detects as evaluation
- Set parentSpanId to the span being evaluated
- If using an LLM for evaluation, include gen_ai.* attributes (will be typed as 'llm')
- Recommended attributes (latest OTel GenAI eval semconv):
  - gen_ai.evaluation.name: name of the evaluation (e.g., "relevance", "toxicity")
  - gen_ai.evaluation.score.value: numeric score (e.g., 0-1 or 1-5)
  - gen_ai.evaluation.score.label: human-readable label (e.g., "pass", "relevant")
  - gen_ai.evaluation.explanation: explanation text from the evaluator
  - (legacy eval.name / eval.score / eval.passed / eval.reason are still accepted)

## Business Attributes
Include custom attributes that identify the business entities your agent works with. Any attributes you add are preserved in metadata and can be queried in TwoTail. Examples:
- user.id, account.id, gen_ai.conversation.id (conversation/session id for chatbots and multi-turn agents)
- document.id, document.url (for RAG agents)
- target.url (for web browsing/fetch agents)
- repo.name, pull_request.id (for coding agents)
- task.type, workflow.name (for productivity and automation agents)

## OTel OTLP/JSON Format
{
  "resourceSpans": [{
    "resource": {
      "attributes": [
        {"key": "service.name", "value": {"stringValue": "my-agent"}}
      ]
    },
    "scopeSpans": [{
      "spans": [{
        "traceId": "abc123...",
        "spanId": "def456...",
        "parentSpanId": null,
        "name": "chat_completion",
        "startTimeUnixNano": 1234567890000000000,
        "endTimeUnixNano": 1234567890100000000,
        "attributes": [
          {"key": "gen_ai.system", "value": {"stringValue": "openai"}},
          {"key": "gen_ai.request.model", "value": {"stringValue": "gpt-4"}},
          {"key": "gen_ai.usage.input_tokens", "value": {"intValue": 150}},
          {"key": "gen_ai.usage.output_tokens", "value": {"intValue": 50}}
        ],
        "status": {"code": 1}
      }]
    }]
  }]
}

## Implementation Notes
- status.code: 0=UNSET, 1=OK, 2=ERROR
- Timestamps are integer Unix time in NANOseconds
- Batch spans and send at most 1,000 per request
- Handle HTTP errors gracefully (don't break the agent if tracing fails)