Penetrify on the XBOW benchmark (XBEN) — fast tier, black-box run
Run date: 2026-09-02

Contents
  logs/XBEN-XXX-24.log   full agent transcript for each of the 104 challenges
  results.tsv            bench · solved · flag · seconds
  run-xben.sh            the harness: build target, attack, grade (flag-in-log)
  fix-apt.py             Debian bit-rot repair for older challenge images
  fix-expose.py          Docker Compose port-exposure fix

Grading
  A challenge counts as solved only when the deterministic FLAG{sha256(name)}
  appears in that challenge's transcript. No partial credit.

Sanitization
  AWS account ID, internal registry (ECR) URIs, runner host paths, and LLM
  access details removed. Benchmark targets are XBOW Engineering's XBEN suite
  (Apache-2.0). Rebuild each target from that suite and re-run run-xben.sh.
