docs/master_data_inventory.md

# Master Data Inventory (v2.0.0 Dataset Adapters)

This inventory tracks **what is implemented in this package** vs. **what is planned** for future adapter expansion.

## Core (Redistributable)

| Dataset | Surface | License | Access | Adapter | Output Included |
| --- | --- | --- | --- | --- | --- |
| HarmBench | B3 | MIT | Manual clone (CSV) | `scripts/adapters/b3_harmbench_adapter.py` | `data/external_augmented/core/B3_augmented_harmbench.jsonl` |
| JailbreakBench (JBB-Behaviors) | B3 | MIT | HuggingFace auto-download | `scripts/adapters/b3_jailbreakbench_adapter.py` | `data/external_augmented/core/B3_augmented_jailbreakbench.jsonl` |
| Gandalf | B3 | MIT | HuggingFace auto-download | `scripts/adapters/b3_gandalf_adapter.py` | `data/external_augmented/core/B3_augmented_gandalf.jsonl` |
| InjecAgent | B4 | MIT | Manual clone (`data/`) | `scripts/adapters/b4_injecagent_adapter.py` | `data/external_augmented/core/B4_augmented_injecagent.jsonl` |
| RAGuard | B1 | MIT | HuggingFace auto-download | `scripts/adapters/b1_raguard_adapter.py` | `data/external_augmented/core/B1_augmented_raguard.jsonl` |
| TensorTrust | B3 | BSD-2-Clause | Manual export (JSONL) | `scripts/adapters/b3_tensortrust_adapter.py` | No (requires user export) |
| ToolEmu | B4 | Apache-2.0 | Manual clone (`assets/all_cases.json`) | `scripts/adapters/b4_toolemu_adapter.py` | `data/external_augmented/core/B4_augmented_toolemu.jsonl` |
| FigStep (SafeBench) | B2 | MIT | Manual clone (images/CSV) | `scripts/adapters/b2_figstep_adapter.py` | `data/external_augmented/core/B2_augmented_figstep.jsonl` (+images) |

## Extended (Research Only / Do Not Redistribute)

| Dataset | Surface | License | Access | Adapter | Output Included |
| --- | --- | --- | --- | --- | --- |
| AgentHarm | B4 | HF “other” (research-use clause) | HuggingFace auto-download | `scripts/adapters/b4_agentharm_adapter.py` | No (generated at runtime) |
| Mindgard | B3 | CC-BY-NC-4.0 (gated) | HuggingFace (requires access) | `scripts/adapters/b3_mindgard_adapter.py` | No (generated at runtime) |
| MM-SafetyBench | B2 | CC-BY-NC-4.0 | HuggingFace auto-download | `scripts/adapters/b2_mmsafetybench_adapter.py` | No (generated at runtime) |