The dataset viewer is not available for this subset.
Exception: SplitsNotFoundError
Message: The split names could not be parsed from the dataset config.
Traceback: Traceback (most recent call last):
File "/usr/local/lib/python3.14/site-packages/datasets/inspect.py", line 286, in get_dataset_config_info
for split_generator in builder._split_generators(
~~~~~~~~~~~~~~~~~~~~~~~~~^
StreamingDownloadManager(base_path=builder.base_path, download_config=download_config)
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
)
^
File "/usr/local/lib/python3.14/site-packages/datasets/packaged_modules/json/json.py", line 101, in _split_generators
pa_table = next(iter(self._generate_tables(**splits[0].gen_kwargs, allow_full_read=False)))[1]
~~~~^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
StopIteration
The above exception was the direct cause of the following exception:
Traceback (most recent call last):
File "/src/services/worker/src/worker/job_runners/config/split_names.py", line 68, in compute_split_names_from_streaming_response
for split in get_dataset_split_names(
~~~~~~~~~~~~~~~~~~~~~~~^
path=dataset,
^^^^^^^^^^^^^
config_name=config,
^^^^^^^^^^^^^^^^^^^
token=hf_token,
^^^^^^^^^^^^^^^
)
^
File "/usr/local/lib/python3.14/site-packages/datasets/inspect.py", line 340, in get_dataset_split_names
info = get_dataset_config_info(
path,
...<6 lines>...
**config_kwargs,
)
File "/usr/local/lib/python3.14/site-packages/datasets/inspect.py", line 291, in get_dataset_config_info
raise SplitsNotFoundError("The split names could not be parsed from the dataset config.") from err
datasets.inspect.SplitsNotFoundError: The split names could not be parsed from the dataset config.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.
Cline Incident Traces
Real coding-agent sessions in which the agent did something unsafe, adversarial, or concerning, contributed by the people who were sitting in front of it. Each session is a redacted transcript, an ATIF trajectory, a structured write-up by the contributor, and, when licensing allows, a snapshot of the repository at the start of the session so the situation can be reproduced.
The dataset exists so that the people training and evaluating models have access to real examples of agents going wrong on real repositories, not only curated benchmarks. Contributions come through the /upload-hf Cline plugin as pull requests to this repository, and every one is reviewed before merge.
What is in a session
sessions/<session_hash>/
SUBMISSION.md contributor's write-up: categories, severity, what happened, where to look
messages.json redacted native transcript (Cline messages contract v1)
trajectory.json ATIF-v1.7 trajectory with outcome signals
meta.json model, provider, repository URL and license, outcome, redaction summary
findings.json redaction counts by detector (never values)
solution.patch the agent's changes as a unified diff, when there were any
repo.tar.gz repository at session start, only for permissive public repositories
manifest.jsonl one line per session for filtering by category, severity, model, repository
SUBMISSION.md is the entry point. Its frontmatter carries the labels (see TAXONOMY.md) and key_messages, the message indices where the incident happens. Its body is the contributor's account in five fixed sections: what they asked for, what the agent did, what they expected, why they are contributing, and notes for reviewers.
session_hash is a truncated SHA-256 of the Cline session id. Working paths, process ids, and raw session ids are never included.
Using the data
Filter manifest.jsonl first. Every line has categories, severity, adversarial, provider, model, repository_license, and snapshot, so you can select, for example, high-severity prompt-injection incidents from a single model family with a snapshot available.
Traces record which model produced each step. Several providers restrict the use of their outputs for training other models. Check the provider's terms before training on a session; the provider and model fields exist so you can build a compliant subset.
How sessions were redacted
Each session went through a deterministic pass (known secrets, credential patterns, home paths, e-mail addresses, contributor-supplied literals and deny-list regexes), a full read-through by the contributor's own agent with a written verdict, and the contributor's explicit confirmation. Sessions were then scanned again during review. If you find something that should not be here, see below.
Contributing
Install the plugin and run /upload-hf in the session you want to report. The agent walks you through redaction review, the write-up, and the pull request. Read CONTRIBUTING.md first, in particular the section on what not to submit.
Removal
Open a discussion titled Removal request: <session_hash>. We remove the session, rebuild the manifest, and request a history purge from Hugging Face when the request involves sensitive data.
License
Transcripts, trajectories, and write-ups are licensed under CC-BY-4.0 by their contributors. Source code inside repo.tar.gz and solution.patch keeps the license of its repository, recorded in meta.json as repository_license.
- Downloads last month
- 50