linuxarena
linuxarena-public
LinuxArena Public Mirror
22,215 agent trajectories from 172 evaluation runs across 10 Linux software environments. Each trajectory carries full per-action content: tool calls, tool outputs, agent reasoning, monitor scores with reasoning and ensemble breakdowns, and blue-protocol audit trails.
Browse interactively: data.linuxarena.ai/datasets
Reviewer sample (data/sample.jsonl, 932 trajs, 705 MB)
Same JSONL schema as the full shards. Each listed run is included in… See the full description on the dataset page: https://huggingface.co/datasets/anonymouslinuxarena/linuxarena-public.linux_arena_goldset-monitor-labelslinuxarena-trajectories
LinuxArena Trajectories
Full agent trajectories from LinuxBench/LinuxArena evaluations
across 14 model/policy combinations and 10 environments.
Dataset Description
Each row is one complete evaluation trajectory — every tool call the agent made from start
to finish, with arguments, outputs, errors, and reasoning. Actions are represented as
parallel variable-length lists (one element per action).
Two granularity levels are provided per action:
Normalized… See the full description on the dataset page: https://huggingface.co/datasets/eval-aware/linuxarena-trajectories.straj_linuxarena
straj_linuxarena (public-env subset)
Adversarial sabotage agentic SWE benchmark trajectories from the linuxarena project, run as part of the no-CoT time-horizons paper.
The dataset viewer above shows the per-cell outcome records (precomputed_results.csv, 257 rows). Full Inspect .eval trajectories for the 13 public-environment tasks (~2.4 GB, 150 files) are stored under evals/ — see "Inspect trajectories" below.
Per-cell schema
Each row in precomputed_results.csv (and… See the full description on the dataset page: https://huggingface.co/datasets/anonymouslinuxarena/straj_linuxarena.linuxarena-first5-trajectories
LinuxArena First-5 Trajectories
First-5 tool call trajectories from LinuxBench/LinuxArena evaluations across 14 model/policy combinations and 10 environments.
Dataset Description
Each row represents one evaluation trajectory with the first 5 tool calls extracted at two granularity levels:
Level 1 (normalized): Tool categories like text_editor:view, bash:find, bash:ls
Level 2 (exact): Full command strings like bash$ find /app/src -type f -name "*.ts" | sort
Designed for… See the full description on the dataset page: https://huggingface.co/datasets/eval-aware/linuxarena-first5-trajectories.linux_arena_preprocessed