WhitzardAgent/AgentTrove-OpenHands

AgentTrove in OpenHands Native trajectory format

About AgentIR Collection

This dataset is part of the [AgentIR Collection](https://huggingface.co/WhitzardAgent). AgentIR is an open-source compiler infrastructure for agentic trajectories (like LLVM/MLIR, but for agent traces). Using AgentIR, you can convert any source trajectory format into multiple target formats.

• —Project: https://github.com/ravenSanstete/agentir
• —DSL: Define custom formats with *.agentir.yaml files
• —CLI: agentir dsl convert for one-command format conversion

Dataset Description

• —Source dataset: open-thoughts/AgentTrove
• —Target format: OpenHands Native trajectory format
• —Rows: 50,000
• —License: apache-2.0

Format Details

Structured trajectory array with action/observation, plus model_patch extracted from diff artifacts.

Record Structure

Each record contains a trajectory field with the converted trajectory data.

{
  "trajectory_id": "rec-001",
  "trajectory": [
    {"role": "user", "content": "Fix the bug"},
    {"role": "assistant", "content": null, "tool_calls": [...]},
    {"role": "tool", "content": "...", "tool_call_id": "call_1"}
  ],
  "model_patch": "diff --git ..."
}

Conversion

This dataset was generated using AgentIR v0.1.0:

# Step 1: Convert source to AgentIR Canonical
agentir dsl convert dsl/formats/agenttrove.agentir.yaml data.jsonl -o canonical.air.jsonl

# Step 2: Lower to target format
agentir-llc --input canonical.air.jsonl --target openhands --output output.jsonl

Pass Pipeline

parse-sharegpt, canonicalize-tools, pair-tool-results, normalize-outcome, verify

Conversion Statistics

Usage

from datasets import load_dataset

ds = load_dataset("WhitzardAgent/AgentTrove-OpenHands", split="train")
print(ds[0])

Re-convert to Other Formats

Since this is an AgentIR-formatted dataset, you can re-convert it to any other target:

# Convert to OpenAI format
agentir-llc --input canonical.air.jsonl --target openai-tools --output openai.jsonl

# Convert to Anthropic format
agentir-llc --input canonical.air.jsonl --target anthropic-tools --output anthropic.jsonl

# Convert to Hermes XML
agentir-llc --input canonical.air.jsonl --target hermes-xml --output hermes.jsonl --include-reasoning

Quality Verification

All conversions passed automated verification:

1. 1.Source dataset field completeness check
2. 2.Role mapping correctness verification (user/assistant/tool/system)
3. 3.Event type inference accuracy verification
4. 4.Zero record errors (100% success rate)
5. 5.Schema compliance validated with agentir verify

Citation

If you use this dataset, please cite both the original source and AgentIR:

@software{agentir,
  title = {AgentIR: A Compiler Infrastructure for Agentic Trajectories},
  url = {https://github.com/ravenSanstete/agentir},
  year = {2026}
}

Generated by AgentIR v0.1.0