armand0e/traces-preview-test

Kimi K2.5 Coding Agent

This is an agentic coding dataset generated by agentic-datagen.

This dataset currently contains 115 rows. The optional preview split contains 115 raw trace files for browsing on the Hub.

Formatting guide

Each row in the dataset includes the same top-level keys: prompt, tools, messages, metadata, and usage.

Completed row without tool use

{
  "prompt": "...",
  "tools": [],
  "messages": [
    {
      "role": "system",
      "content": "You are a coding agent."
    },
    {
      "role": "user",
      "content": "..."
    },
    {
      "role": "assistant",
      "reasoning_content": "...",
      "content": "Final answer..."
    }
  ],
  "metadata": {
    "prompt_id": "prompt_000000_abc123",
    "run_id": "run_20260414T000000Z",
    "session_id": "session_000000",
    "turns": 1,
    "completed": true,
    "tool_calls_count": 0,
    "error": null,
    "retryable": false
  },
  "usage": {
    "prompt_tokens": 256,
    "completion_tokens": 128,
    "reasoning_tokens": 32,
    "total_tokens": 416,
    "cost": 0.0
  }
}

Completed row with tool use

{
  "prompt": "...",
  "tools": [
    {
      "type": "function",
      "function": {
        "name": "write_file",
        "description": "Create or overwrite a file in the workspace.",
        "parameters": {
          "type": "object",
          "properties": {
            "file_path": {
              "type": "string"
            },
            "content": {
              "type": "string"
            }
          },
          "required": [
            "file_path",
            "content"
          ]
        }
      }
    }
  ],
  "messages": [
    {
      "role": "system",
      "content": "You are a coding agent."
    },
    {
      "role": "user",
      "content": "..."
    },
    {
      "role": "assistant",
      "reasoning_content": "...",
      "content": "",
      "tool_calls": [
        {
          "id": "call_1",
          "type": "function",
          "index": 0,
          "function": {
            "name": "write_file",
            "arguments": {
              "file_path": "...",
              "content": "..."
            }
          }
        }
      ]
    },
    {
      "role": "tool",
      "tool_call_id": "call_1",
      "name": "write_file",
      "content": "..."
    },
    {
      "role": "assistant",
      "content": "Final answer after tool result..."
    }
  ],
  "metadata": {
    "prompt_id": "prompt_000000_abc123",
    "run_id": "run_20260414T000000Z",
    "session_id": "session_000000",
    "turns": 2,
    "completed": true,
    "tool_calls_count": 1,
    "error": null,
    "retryable": false
  },
  "usage": {
    "prompt_tokens": 512,
    "completion_tokens": 160,
    "reasoning_tokens": 48,
    "total_tokens": 720,
    "cost": 0.0
  }
}

reasoning_content is only included when a reasoning trace is present. Assistant tool_calls stay inside the assistant message, tool responses stay in tool messages, and the top-level tools, metadata, and usage fields are preserved on every row.

Preview split trace file

When a preview directory is configured, the generator also emits one JSONL trace file per session. Each file contains newline-delimited raw trace events similar to the agent-export format used by Hugging Face trace datasets.

{"type": "session", "version": 3, "id": "session_000000", "timestamp": "2026-04-14T00:00:00.000Z", "cwd": "/workspace/session_000000"}
{"type": "model_change", "id": "session_000000_model", "parentId": null, "timestamp": "2026-04-14T00:00:00.001Z", "provider": "openrouter", "modelId": "example-model"}
{"type": "session_info", "id": "session_000000_info", "parentId": "session_000000_model", "timestamp": "2026-04-14T00:00:00.002Z", "name": "..."}

Generation metadata

• Model: not set
• Reasoning effort: not set
• Enabled tools: No tools configured
• Row metadata fields: prompt_id, run_id, session_id, turns, completed, tool_calls_count, error, retryable.
• Usage fields: prompt_tokens, completion_tokens, reasoning_tokens, total_tokens, cost.