skillware-bench 0.3.2__py3-none-any.whl

This diff represents the content of publicly available package versions that have been released to one of the supported registries. The information contained in this diff is provided for informational purposes only and reflects changes between package versions as they appear in their respective public registries.
Files changed (49) hide show
  1. skillware_bench/__init__.py +62 -0
  2. skillware_bench/cli.py +639 -0
  3. skillware_bench/core/adapters/__init__.py +50 -0
  4. skillware_bench/core/adapters/anthropic_adapter.py +166 -0
  5. skillware_bench/core/adapters/base.py +182 -0
  6. skillware_bench/core/adapters/factory.py +140 -0
  7. skillware_bench/core/adapters/gemini_adapter.py +179 -0
  8. skillware_bench/core/adapters/mistral_adapter.py +136 -0
  9. skillware_bench/core/adapters/mock_adapter.py +100 -0
  10. skillware_bench/core/adapters/ollama_adapter.py +137 -0
  11. skillware_bench/core/adapters/openai_adapter.py +164 -0
  12. skillware_bench/core/adapters/openrouter_adapter.py +446 -0
  13. skillware_bench/core/adapters/registry.py +8 -0
  14. skillware_bench/core/agent_loop.py +383 -0
  15. skillware_bench/core/evaluators.py +246 -0
  16. skillware_bench/core/models.py +96 -0
  17. skillware_bench/core/openrouter_engine.py +357 -0
  18. skillware_bench/core/perturbations.py +130 -0
  19. skillware_bench/core/provenance.py +371 -0
  20. skillware_bench/core/runner.py +147 -0
  21. skillware_bench/core/statistics.py +123 -0
  22. skillware_bench/dashboard/__init__.py +8 -0
  23. skillware_bench/dashboard/generator.py +248 -0
  24. skillware_bench/fixtures/__init__.py +9 -0
  25. skillware_bench/fixtures/mock_companies_registry.py +419 -0
  26. skillware_bench/fixtures/mock_inbox_registry.py +228 -0
  27. skillware_bench/infra/__init__.py +27 -0
  28. skillware_bench/infra/companies_house_live.py +228 -0
  29. skillware_bench/infra/disposable_mailbox.py +210 -0
  30. skillware_bench/infra/evm_testnet_fork.py +179 -0
  31. skillware_bench/infra/hermetic_mock_server.py +354 -0
  32. skillware_bench/infra/wallet_screening_live.py +143 -0
  33. skillware_bench/suites/__init__.py +33 -0
  34. skillware_bench/suites/chained_pipeline_crucible.py +177 -0
  35. skillware_bench/suites/companies_house_crucible.py +187 -0
  36. skillware_bench/suites/constitution_crucible.py +171 -0
  37. skillware_bench/suites/context_drift_crucible.py +254 -0
  38. skillware_bench/suites/distraction_registry.py +249 -0
  39. skillware_bench/suites/gmail_crucible.py +207 -0
  40. skillware_bench/suites/prompt_injection_crucible.py +263 -0
  41. skillware_bench/suites/wallet_screening_crucible.py +167 -0
  42. skillware_bench/viz/__init__.py +8 -0
  43. skillware_bench/viz/generator.py +179 -0
  44. skillware_bench-0.3.2.dist-info/METADATA +287 -0
  45. skillware_bench-0.3.2.dist-info/RECORD +49 -0
  46. skillware_bench-0.3.2.dist-info/WHEEL +5 -0
  47. skillware_bench-0.3.2.dist-info/entry_points.txt +2 -0
  48. skillware_bench-0.3.2.dist-info/licenses/LICENSE +21 -0
  49. skillware_bench-0.3.2.dist-info/top_level.txt +1 -0
@@ -0,0 +1,62 @@
1
+ # Copyright (c) 2026 ARPA Hellenic Logical Systems
2
+ # SPDX-License-Identifier: MIT
3
+
4
+ """Skillware-Bench: The Empirical Benchmark & Evaluation Harness for Governed Agent Skills."""
5
+
6
+ import os
7
+ import sys
8
+
9
+
10
+ def _sanitize_import_paths() -> None:
11
+ """Ensure installed packages take precedence over local reference checkout directories."""
12
+ cwd = os.getcwd()
13
+ for p in ["", cwd]:
14
+ if p in sys.path:
15
+ cand = os.path.join(p, "skillware")
16
+ if os.path.isdir(cand) and not os.path.isfile(os.path.join(cand, "__init__.py")):
17
+ try:
18
+ sys.path.remove(p)
19
+ except ValueError:
20
+ pass
21
+
22
+
23
+ _sanitize_import_paths()
24
+
25
+ from skillware_bench.core.adapters import ( # noqa: E402
26
+ AdapterRegistry,
27
+ BaseModelAdapter,
28
+ ModelResponse,
29
+ TokenUsage,
30
+ get_adapter,
31
+ )
32
+ from skillware_bench.core.agent_loop import AgentLoop # noqa: E402
33
+ from skillware_bench.core.evaluators import CrucibleEvaluator # noqa: E402
34
+ from skillware_bench.core.perturbations import PerturbationEngine # noqa: E402
35
+ from skillware_bench.core.provenance import ProvenanceEngine # noqa: E402
36
+ from skillware_bench.core.runner import HarnessRunner # noqa: E402
37
+ from skillware_bench.core.statistics import StatisticalEvaluator # noqa: E402
38
+ from skillware_bench.dashboard import DashboardGenerator # noqa: E402
39
+ from skillware_bench.suites import ALL_CRUCIBLES # noqa: E402
40
+ from skillware_bench.viz import VisualGenerator # noqa: E402
41
+
42
+ __version__ = "0.3.2"
43
+ __author__ = "ARPA Hellenic Logical Systems"
44
+
45
+ __all__ = [
46
+ "__version__",
47
+ "__author__",
48
+ "HarnessRunner",
49
+ "AgentLoop",
50
+ "CrucibleEvaluator",
51
+ "ProvenanceEngine",
52
+ "PerturbationEngine",
53
+ "StatisticalEvaluator",
54
+ "DashboardGenerator",
55
+ "VisualGenerator",
56
+ "AdapterRegistry",
57
+ "BaseModelAdapter",
58
+ "ModelResponse",
59
+ "TokenUsage",
60
+ "get_adapter",
61
+ "ALL_CRUCIBLES",
62
+ ]