Cybersecurity reasoning tasks pairing vulnerability analysis with secure patches, exploit evidence, and regression validation at repository scale. Measured against CVE-Bench and CyberGym targets.
Available Now · Sample Available
Cybersecurity Patch Dataset
Repository-scale patching and exploit-trajectory tasks with executable validation.
Volume
0
Volume
Scalable Volume
Coverage
Gold Harness Validation
0.0
%
Gold reference trajectories
Targets
CVE-Bench, CyberGym
Modality & Delivery
TextCodeRepository→CodeTextPatch
Formats
RepositoryJSON
What You Receive
Coverage
A cybersecurity reasoning suite spanning vulnerability remediation, inverse discovery, supply-chain and cloud security, CVE/CWE mapping, detection, exploit reproduction, secure patching, and regression validation.
Package Structure
- trajectory.jsonNormalized vulnerability, repository, patch, exploit-reasoning, verification, and reward record.
- patch.diffReference source-code patch in unified diff format.
- exploit/Runnable exploit or proof-of-concept harness.
- environment/Reproducible build and execution configuration.
- results/Pre-patch and post-patch execution evidence.
- analysis/Optional program-analysis and graph artifacts.
- provenance/Manifest, quality, coverage, lineage, and checksum metadata.
Differentiators
Record Schema
Results
| Subject | Type | Metric | Result |
|---|---|---|---|
| Gold reference trajectories | Gold validation | Pass rate |
Methodology & Quality
Methodology
Task families retain repository context, vulnerability metadata, patch artifacts, exploit trajectories, and validation outputs where applicable.
SME involvement
Domain-expert curated with up to L2 review
Stressors
Repository navigation, vulnerable-path discovery, CVE/CWE reasoning, exploit reproduction, secure patch generation, regression preservation, and software-supply-chain dependencies.
Design criteria
- Require repository-level evidence for vulnerability diagnosis and remediation.
- Validate that a secure patch blocks the exploit without breaking intended behavior.
Build assumptions
Quality approach
Known limitations
Customization & Scalability
Custom sets available. Additional volume in progress