unix-ctf: Procedural Environments for Unix-Competence Reinforcement Learning

Geoffrey Bradway, Roger Creus Castanyer, Lorenz Wolf, Maxwill Lin, Matthew James Sargent, Augustine N. Mavor-Parker · 2026

Read unix-ctf: Procedural Environments for Unix-Competence Reinforcement Learning by Geoffrey Bradway, Roger Creus Castanyer, Lorenz Wolf, Maxwill Lin, Matthew James Sargent, Augustine N. Mavor-Parker (2026) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2605-29115.txt.

Description

Geoffrey Bradway and colleagues build unix-ctf, a generator of capture-the-flag tasks. Each task hides a flag in a fresh Linux container using one Unix feature, and the agent recovers it. An LLM-assisted pipeline writes planting and recovery scripts under a two-way contract. Hide scripts leave no plaintext flag trace, and find scripts recover the flag in a fresh directory. The pipeline lands 656 of 750 as variants covering 155 techniques, far above baseline. Qwen3-8B fine-tuning lifts solve rates from about 12 to about 44 percent.

Catalog record

CREATORGeoffrey Bradway, Roger Creus Castanyer, Lorenz Wolf, Maxwill Lin, Matthew James Sargent, Augustine N. Mavor-Parker
YEAR2026
COLLECTIONPAPERS
FORMATTXT
FILE SIZE66216
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2605.29115

Browse the full library · About the Tower of Babel