Specific capability target
The task is built around a named security skill rather than a broad theme or an arbitrary puzzle mechanic.
ORIGINAL CYBERSECURITY TASKS
Original cybersecurity tasks, built and verified by experts.
CYVARC creates new tasks around the security capability your model needs to learn or prove. We define the objective, environment, constraints, intended solution, and validation method as one system. Every task is reviewed by people who work in the relevant security domain.
What makes a task ready for training.
A CTF challenge can be a useful starting point, but training data needs more structure. The objective, environment, solution, and reward must remain consistent when another expert or agent approaches the task.
Reproducible environment
The setup is packaged so reviewers and agents encounter the same conditions and the result does not depend on an accidental local state.
Independent reproduction
A second specialist follows the task from the supplied materials and verifies the intended solution without relying on private author context.
From specification to reviewed task.
Define the objective
We agree on the target skill, expected agent access, completion boundary, difficulty range, and any restrictions before authoring begins.
Build the task and environment
The author creates the challenge materials, setup, intended path, and checks required to reproduce the result.
Review what the task actually measures
An expert checks whether the task rewards the intended capability and flags ambiguity, unintended paths, fragile dependencies, or difficulty caused by poor setup.
Package the evidence
Deliverables can include task files, environment instructions, the verified solution, agent run notes, reward checks, and structured review findings.
Tell us what your model needs to learn or prove.
Share the target capability, domain, expected volume, agent setup, delivery format, and review requirements. We will shape the task and validation plan around them.