README.md

metadata

pretty_name: OpenToolTrace-X (Platinum)
language:
  - en
license: apache-2.0
task_categories:
  - text-generation
  - question-answering
  - reinforcement-learning
tags:
  - agents
  - tool-use
  - trajectories
  - verification
  - code
  - bash
  - git
size_categories:
  - n<1K
dataset_info:
  creator: Within US AI
  contact: Within US AI
  created: '2025-12-30T16:53:41Z'
  schema: See Features section below

OpenToolTrace-X (Platinum)

Developer/Publisher: Within US AI
Version: 0.1.0 (sample pack)
Created: 2025-12-30T16:53:41Z

What this dataset is

OpenToolTrace-X is a replayable, verifiable corpus of tool-using agent trajectories.

Each record contains:

• A user goal (prompt) and constraints
• An initial_state describing the starting environment/repo snapshot
• A trajectory (tool calls + observations)
• A final_state (artifacts/diff/output)
• verification (tests, checksums, exit codes) to make outcomes machine-checkable

Features / schema (JSONL)

• task_id (string)
• domain (string; e.g., python, bash, git, data)
• difficulty (int; 1–5)
• prompt (string)
• constraints (string)
• initial_state (object)
• trajectory (list of objects)
• final_state (object)
• verification (object)
• tags (list of strings)
• created_utc (string; ISO 8601)
• license_note (string)

Trajectory step format

Each step is a dict:

• tool (e.g., bash, python, git)
• action (command / code / args)
• observation (stdout / structured output)
• exit_code (int)
• stderr (string, optional)
• artifacts_written (list of strings, optional)

Data splits

• data/train.jsonl
• data/validation.jsonl
• data/test.jsonl

Replay harness (scaffold)

See replay_harness/ for a safe, non-executing replay viewer. Integrate your own sandbox executor for real replays.

How to load

from datasets import load_dataset
ds = load_dataset("json", data_files={
    "train": "data/train.jsonl",
    "validation": "data/validation.jsonl",
    "test": "data/test.jsonl",
})
print(ds["train"][0])