| [project] | |
| name = "side-channel-testbed" | |
| version = "0.1.0" | |
| description = "Replication package: behavioural side-channel inference testbed for browser-use agents" | |
| requires-python = ">=3.13" | |
| dependencies = [ | |
| "browser-use[core]>=0.13.1", | |
| "fastapi>=0.136.3", | |
| "jinja2>=3.1.6", | |
| "matplotlib>=3.11.0", | |
| "pandas>=3.0.3", | |
| "pytest>=9.0.3", | |
| "python-dotenv>=1.2.2", | |
| "scikit-learn>=1.9.0", | |
| "scipy>=1.17.1", | |
| "tqdm>=4.66", | |
| "uvicorn>=0.49.0", | |
| "httpx>=0.27.0", | |
| "pyyaml>=6.0.3", | |
| ] | |
| [tool.ruff] | |
| # The code was written to a 100-column limit, not Ruff's default of 88. | |
| line-length = 100 | |
| target-version = "py313" | |
| [tool.ruff.lint] | |
| select = ["E", "W", "F", "B", "C4", "N", "I", "S", "UP"] | |
| [tool.ruff.lint.per-file-ignores] | |
| # The scripts are local entry points: they shell out to fixed commands and fetch loopback | |
| # URLs by design, and they import after inserting the repo root on sys.path. | |
| "scripts/*" = ["S603", "S310", "E402"] | |
| # Same for the report exporter, which re-invokes this project per dataset. | |
| "analysis/export_report.py" = ["S603"] | |
| # P, Ph and Ps are the probability matrix and its marginals from the mutual-information | |
| # definition in plan.md, and match the notation in the surrounding docstring. | |
| "analysis/channel_metrics.py" = ["N806"] | |
| # Per-run option-order randomization and target rotation are seeded experimental controls | |
| # (MATRIX.base_seed), not security primitives. Swapping `random` for `secrets` would make | |
| # every recorded run irreproducible. | |
| "attacker/app.py" = ["S311"] | |