# WWT Demo — Public Agentic-Security Dataset Archives

- Date: 2026-07-07
- Owner: Nick Venezia / Centillion.AI
- R2 bucket: `trustfortress-resource-library`
- Library prefix: `cubie-tf/wwt-demo-assets/datasets/2026-07-07/`
- Object route: `https://lib.trustfortress.ai/objects/<url-encoded-key>`
- Browse: `https://lib.trustfortress.ai/api/list?prefix=cubie-tf%2Fwwt-demo-assets%2Fdatasets%2F2026-07-07%2F`
- Access class: **public** (all four are already-public OSS upstream; archived here for one-stop launch-day access)

Reference-material archive only. These are `git clone --depth 1` snapshots of four
public agentic-security repositories, tarred and stored for the WWT demo. **No
dataset code is imported into the cubie-tf codebase** (repo no-fork / clean-code
rule) — this is a byte-for-byte upstream snapshot for provenance, not a fork.

## Datasets

| Dataset | Upstream | License | Commit (resolved HEAD) | LICENSE file | Clone size | Archive | Archive bytes | Archive SHA-256 |
|---|---|---|---|---|---:|---|---:|---|
| finbot-ctf | https://github.com/OWASP-ASI/finbot-ctf | Apache-2.0 | `1450fc4d15cbe80dbaf52dde1df767dbc967e32e` | `LICENSE.md` | 25M | `finbot-ctf-1450fc4.tar.gz` | 21215274 | `52182f644dd7cd670fccfe74ef57c8880ef9a6329a3347234ca1f49116ebc2d9` |
| skillguard | https://github.com/obielin/skillguard | MIT | `4df364544fad70c50978c9cc97e2871f0aa3ed47` | `LICENSE` | 203K | `skillguard-4df3645.tar.gz` | 58328 | `784800bc7a0990d602201e83e94442fe0de8b8a18c6cbf3b365501879c0c3d0f` |
| atlas-data | https://github.com/mitre-atlas/atlas-data | Apache-2.0 (MITRE open data) | `651dad90d3c007e797c89356fa1f4d8732f90c8d` | `LICENSE` | 20M | `atlas-data-651dad9.tar.gz` | 5828508 | `77f466ccc0165b3cacd8f40e30e5d90bfeffa98dadecf68b0b31bf516a5860d3` |
| agentdojo-inspect | https://github.com/usnistgov/agentdojo-inspect | NIST public-service software (US public domain; upstream AgentDojo MIT) | `a491020fd14e87bcc19b2eb77fe307d8b329b05d` | `LICENSE.md` | 300M | `agentdojo-inspect-a491020f.tar.gz` | 47733038 | `c5bdc4a693265d9cb731040bae635e1e976e10978204fa99382dd877f0a50f8e` |

### What each is

- **finbot-ctf** (OWASP-ASI, Apache-2.0) — multi-agent fintech red-team / capture-the-flag
  sandbox exercising agentic + MCP tool misuse.
- **skillguard** (obielin, MIT) — static agent-skill / MCP scanner mapped to OWASP AST10.
- **atlas-data** (MITRE ATLAS, Apache-2.0) — MITRE ATLAS adversarial-ML threat YAML.
  Threat taxonomy at `dist/ATLAS.yaml` (`dist/ATLAS-latest.yaml` is a text pointer to
  `v6/ATLAS-latest.yaml`; both the pinned and versioned files are inside the archive).
- **agentdojo-inspect** (usnistgov, NIST software license) — indirect-prompt-injection (IPI)
  and RCE benchmark; UK AISI / NIST Inspect port of AgentDojo (upstream AgentDojo is MIT).

## Secret scan (pre-upload)

Every cloned file was scanned for real-credential markers
(`sk-ant-…`, `AKIA[0-9A-Z]{16}`, `glpat-…`, `-----BEGIN … PRIVATE KEY-----`,
and password-assignment literals). **No real credentials.** The three high-signal
hits are all synthetic / test fixtures, safe to archive:

1. `skillguard/tests/test_skillguard.py` → `AKIAIOSFODNN7EXAMPLE` — AWS's official
   public documentation placeholder access key (not a real key).
2. `skillguard/tests/test_skillguard.py` → `-----BEGIN RSA PRIVATE KEY-----` — a
   scanner test assertion (header string only, no key body).
3. `agentdojo-inspect/.../mock-gist-server/certs/nginx.key` — a self-signed mock
   test cert (subject == issuer, placeholder org, generated by `gen-certs.sh`) for a
   localhost benchmark mock server; already public upstream.

## Reproduction

```bash
# fetch an archive from R2 and extract
curl -L "https://lib.trustfortress.ai/objects/cubie-tf%2Fwwt-demo-assets%2Fdatasets%2F2026-07-07%2Fatlas-data-651dad9.tar.gz" -o atlas-data.tar.gz
tar -xzf atlas-data.tar.gz
# or re-clone upstream at the pinned commit
git clone https://github.com/mitre-atlas/atlas-data && git -C atlas-data checkout 651dad90d3c007e797c89356fa1f4d8732f90c8d
```