Hugging Face's logo

Datasets:
Quanyu001
/
trace2env-worldbook-terminal

ArXiv:
Tags:
trace2env
world-model
environment-simulation
terminal
agents
License:
Dataset card Files
xet
 Community
Dataset Viewer
Duplicate
The dataset viewer is not available for this split.
Cannot load the dataset split (in streaming mode) to extract the first rows.
Error code:   StreamingRowsError
Exception:    CastError
Message:      Couldn't cast
id: string
action_type: string
description: string
priority: int64
conditions: list<item: struct<left: struct<path: string, action_arg: string, literal: null>, op: string, right:  (... 59 chars omitted)
  child 0, item: struct<left: struct<path: string, action_arg: string, literal: null>, op: string, right: struct<path (... 47 chars omitted)
      child 0, left: struct<path: string, action_arg: string, literal: null>
          child 0, path: string
          child 1, action_arg: string
          child 2, literal: null
      child 1, op: string
      child 2, right: struct<path: string, action_arg: string, literal: string>
          child 0, path: string
          child 1, action_arg: string
          child 2, literal: string
effects: list<item: struct<op: string, path: string, value: string, delay_steps: null>>
  child 0, item: struct<op: string, path: string, value: string, delay_steps: null>
      child 0, op: string
      child 1, path: string
      child 2, value: string
      child 3, delay_steps: null
outcome: string
renderer: string
observation_template: null
confidence: double
status: string
provenance: list<item: struct<source_id: string, episode_id: string, event_ids: list<item: string>, locator: str (... 22 chars omitted)
  child 0, item: struct<source_id: string, episode_id: string, event_ids: list<item: string>, locator: string, excerp (... 10 chars omitted)
      child 0, source_id: string
      child 1, episode_id: string
      child 2, event_ids: list<item: st
...
 0, type: string
              child 1, required: bool
              child 2, description: string
              child 3, enum: null
          child 5, program: struct<type: string, required: bool, description: string, enum: null>
              child 0, type: string
              child 1, required: bool
              child 2, description: string
              child 3, enum: null
          child 6, command: struct<type: string, required: bool, description: string, enum: null>
              child 0, type: string
              child 1, required: bool
              child 2, description: string
              child 3, enum: null
          child 7, argv: struct<type: string, required: bool, description: string, enum: null>
              child 0, type: string
              child 1, required: bool
              child 2, description: string
              child 3, enum: null
      child 3, aliases: list<item: null>
          child 0, item: null
      child 4, provenance: list<item: struct<source_id: string, episode_id: string, event_ids: list<item: string>, locator: str (... 22 chars omitted)
          child 0, item: struct<source_id: string, episode_id: string, event_ids: list<item: string>, locator: string, excerp (... 10 chars omitted)
              child 0, source_id: string
              child 1, episode_id: string
              child 2, event_ids: list<item: string>
                  child 0, item: string
              child 3, locator: string
              child 4, excerpt: string
to
{'actions': List({'name': Value('string'), 'description': Value('string'), 'arguments': {'commands': {'type': Value('string'), 'required': Value('bool'), 'description': Value('string'), 'enum': Value('null')}, 'programs': {'type': Value('string'), 'required': Value('bool'), 'description': Value('string'), 'enum': Value('null')}, 'keystrokes': {'type': Value('string'), 'required': Value('bool'), 'description': Value('string'), 'enum': Value('null')}, 'duration': {'type': Value('string'), 'required': Value('bool'), 'description': Value('string'), 'enum': Value('null')}, 'keys': {'type': Value('string'), 'required': Value('bool'), 'description': Value('string'), 'enum': Value('null')}, 'program': {'type': Value('string'), 'required': Value('bool'), 'description': Value('string'), 'enum': Value('null')}, 'command': {'type': Value('string'), 'required': Value('bool'), 'description': Value('string'), 'enum': Value('null')}, 'argv': {'type': Value('string'), 'required': Value('bool'), 'description': Value('string'), 'enum': Value('null')}}, 'aliases': List(Value('null')), 'provenance': List({'source_id': Value('string'), 'episode_id': Value('string'), 'event_ids': List(Value('string')), 'locator': Value('string'), 'excerpt': Value('string')})})}
because column names don't match
Traceback:    Traceback (most recent call last):
                File "/src/services/worker/src/worker/utils.py", line 147, in get_rows_or_raise
                  return get_rows(
                      dataset=dataset,
                  ...<4 lines>...
                      column_names=column_names,
                  )
                File "/src/libs/libcommon/src/libcommon/utils.py", line 272, in decorator
                  return func(*args, **kwargs)
                File "/src/services/worker/src/worker/utils.py", line 127, in get_rows
                  rows_plus_one = list(itertools.islice(safe_iter(ds, dataset=dataset), rows_max_number + 1))
                File "/src/services/worker/src/worker/utils.py", line 483, in safe_iter
                  yield from ds.decode(False) if ds.features else ds
                File "/usr/local/lib/python3.14/site-packages/datasets/iterable_dataset.py", line 2951, in __iter__
                  for key, example in ex_iterable:
                                      ^^^^^^^^^^^
                File "/usr/local/lib/python3.14/site-packages/datasets/iterable_dataset.py", line 2461, in __iter__
                  for key, pa_table in self._iter_arrow():
                                       ~~~~~~~~~~~~~~~~^^
                File "/usr/local/lib/python3.14/site-packages/datasets/iterable_dataset.py", line 2486, in _iter_arrow
                  for key, pa_table in self.ex_iterable._iter_arrow():
                                       ~~~~~~~~~~~~~~~~~~~~~~~~~~~~^^
                File "/usr/local/lib/python3.14/site-packages/datasets/iterable_dataset.py", line 547, in _iter_arrow
                  for key, pa_table in iterator:
                                       ^^^^^^^^
                File "/usr/local/lib/python3.14/site-packages/datasets/iterable_dataset.py", line 430, in _iter_arrow
                  for key, pa_table in self.generate_tables_fn(**gen_kwags):
                                       ~~~~~~~~~~~~~~~~~~~~~~~^^^^^^^^^^^^^
                File "/usr/local/lib/python3.14/site-packages/datasets/packaged_modules/json/json.py", line 343, in _generate_tables
                  self._cast_table(pa_table, json_field_paths=json_field_paths),
                  ~~~~~~~~~~~~~~~~^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
                File "/usr/local/lib/python3.14/site-packages/datasets/packaged_modules/json/json.py", line 132, in _cast_table
                  pa_table = table_cast(pa_table, self.info.features.arrow_schema)
                File "/usr/local/lib/python3.14/site-packages/datasets/table.py", line 2378, in table_cast
                  return cast_table_to_schema(table, schema)
                File "/usr/local/lib/python3.14/site-packages/datasets/table.py", line 2306, in cast_table_to_schema
                  raise CastError(
                  ...<3 lines>...
                  )
              datasets.table.CastError: Couldn't cast
              id: string
              action_type: string
              description: string
              priority: int64
              conditions: list<item: struct<left: struct<path: string, action_arg: string, literal: null>, op: string, right:  (... 59 chars omitted)
                child 0, item: struct<left: struct<path: string, action_arg: string, literal: null>, op: string, right: struct<path (... 47 chars omitted)
                    child 0, left: struct<path: string, action_arg: string, literal: null>
                        child 0, path: string
                        child 1, action_arg: string
                        child 2, literal: null
                    child 1, op: string
                    child 2, right: struct<path: string, action_arg: string, literal: string>
                        child 0, path: string
                        child 1, action_arg: string
                        child 2, literal: string
              effects: list<item: struct<op: string, path: string, value: string, delay_steps: null>>
                child 0, item: struct<op: string, path: string, value: string, delay_steps: null>
                    child 0, op: string
                    child 1, path: string
                    child 2, value: string
                    child 3, delay_steps: null
              outcome: string
              renderer: string
              observation_template: null
              confidence: double
              status: string
              provenance: list<item: struct<source_id: string, episode_id: string, event_ids: list<item: string>, locator: str (... 22 chars omitted)
                child 0, item: struct<source_id: string, episode_id: string, event_ids: list<item: string>, locator: string, excerp (... 10 chars omitted)
                    child 0, source_id: string
                    child 1, episode_id: string
                    child 2, event_ids: list<item: st
              ...
               0, type: string
                            child 1, required: bool
                            child 2, description: string
                            child 3, enum: null
                        child 5, program: struct<type: string, required: bool, description: string, enum: null>
                            child 0, type: string
                            child 1, required: bool
                            child 2, description: string
                            child 3, enum: null
                        child 6, command: struct<type: string, required: bool, description: string, enum: null>
                            child 0, type: string
                            child 1, required: bool
                            child 2, description: string
                            child 3, enum: null
                        child 7, argv: struct<type: string, required: bool, description: string, enum: null>
                            child 0, type: string
                            child 1, required: bool
                            child 2, description: string
                            child 3, enum: null
                    child 3, aliases: list<item: null>
                        child 0, item: null
                    child 4, provenance: list<item: struct<source_id: string, episode_id: string, event_ids: list<item: string>, locator: str (... 22 chars omitted)
                        child 0, item: struct<source_id: string, episode_id: string, event_ids: list<item: string>, locator: string, excerp (... 10 chars omitted)
                            child 0, source_id: string
                            child 1, episode_id: string
                            child 2, event_ids: list<item: string>
                                child 0, item: string
                            child 3, locator: string
                            child 4, excerpt: string
              to
              {'actions': List({'name': Value('string'), 'description': Value('string'), 'arguments': {'commands': {'type': Value('string'), 'required': Value('bool'), 'description': Value('string'), 'enum': Value('null')}, 'programs': {'type': Value('string'), 'required': Value('bool'), 'description': Value('string'), 'enum': Value('null')}, 'keystrokes': {'type': Value('string'), 'required': Value('bool'), 'description': Value('string'), 'enum': Value('null')}, 'duration': {'type': Value('string'), 'required': Value('bool'), 'description': Value('string'), 'enum': Value('null')}, 'keys': {'type': Value('string'), 'required': Value('bool'), 'description': Value('string'), 'enum': Value('null')}, 'program': {'type': Value('string'), 'required': Value('bool'), 'description': Value('string'), 'enum': Value('null')}, 'command': {'type': Value('string'), 'required': Value('bool'), 'description': Value('string'), 'enum': Value('null')}, 'argv': {'type': Value('string'), 'required': Value('bool'), 'description': Value('string'), 'enum': Value('null')}}, 'aliases': List(Value('null')), 'provenance': List({'source_id': Value('string'), 'episode_id': Value('string'), 'event_ids': List(Value('string')), 'locator': Value('string'), 'excerpt': Value('string')})})}
              because column names don't match

Need help to make the dataset viewer work? Make sure to review how to configure the dataset viewer, and open a discussion for direct support.

Trace2Env terminal worldbook

The frozen terminal worldbook of the paper From Traces to Agentic Worlds: Agentic Language World Models for Interactive Environment Simulation (arXiv:2610.06100), the one that produced the paper's Terminal results and that the live demo runs on. It is also the quickest way to see what a worldbook looks like and to try the world model agent without building one.

What it is

A Trace2Env worldbook is reusable, inspectable environment knowledge reconstructed from interaction traces: action and state schemas, grounded evidence (observed transitions and demonstrations), induced rules, invariants, observation contracts and notes, each tied to the traces that support it. A world model agent operates it to simulate the environment for a task agent: it inspects the worldbook, keeps the episode's state and memory, and answers each action with the next observation and the state changes that persist. Code: https://github.com/ruyue0001/trace2env.

  • Environment: a Linux shell as seen by a Terminus-2 agent through tmux on Terminal-Bench 2.0 tasks.
  • Construction data: 20 trajectories collected by the authors with a GPT-5.6-Sol task agent on Terminal-Bench 2.0 (334 transitions), chosen from tasks that do not overlap the AgentWorldBench terminal records. The worldbook embeds those runs' observations verbatim as evidence.
  • Reconstruction model: GPT-5.6-Sol, Trace2Env's default pipeline. Validation status candidate (no replay validation was run), so the runtime needs --allow-unvalidated.
  • Integrity: worldbook/manifest.json lists the SHA-256 of every file; Trace2Env re-verifies them on load. Artifact digest 1a3a0bf8…, config digest cc3e9094….

Contents 

worldbook/              the package: action_schema.json, state_schema.json, rules/, invariants.json, renderer/,
                        knowledge/notes.jsonl, evidence/, demonstrations/, candidates/, construction/, index.sqlite,
                        manifest.json, overview.md
environment_prompt.txt  the terminal environment's system prompt (the one the demo passes as --environment-prompt)

Use it 

pip install huggingface_hub
hf download Quanyu001/trace2env-worldbook-terminal --repo-type dataset --local-dir work/worldbooks/terminal

git clone https://github.com/ruyue0001/trace2env.git && cd trace2env && python -m pip install -e .
python -m trace2env inspect ../work/worldbooks/terminal/worldbook --action cat       # what it knows about one action
MODEL=(--provider chat --base-url https://YOUR-ENDPOINT/v1 --api-key-env YOUR_KEY_VAR --model YOUR_MODEL)
python -m trace2env simulate ../work/worldbooks/terminal/worldbook --session work/session-1 --allow-unvalidated --agent "${MODEL[@]}" \
  --initial-state '{"session": {"cwd": "/app"}}' --action '{"type": "shell", "arguments": {"commands": ["ls"]}}'

Each simulate call is one task-agent action; the session keeps state and memory between calls. Any chat model served through an API endpoint can play the world model.

Citation 

@article{long2026agenticworlds,
  title={From Traces to Agentic Worlds: Agentic Language World Models for Interactive Environment Simulation},
  author={Long, Quanyu and Chen, Xiao and Chen, Jianda and Zhang, Haozhen and Hu, Qisheng and Bao, Jianzhu and Wang, Wenya},
  journal={arXiv preprint arXiv:2610.06100},
  year={2026}
}
Downloads last month
171
Total file size:
25.1 MB

Paper for Quanyu001/trace2env-worldbook-terminal