<a name="top"></a>
<div align="center">

<img src="https://capsule-render.vercel.app/api?type=rect&color=0:6b46c1,100:2b6cb0&height=120&section=header&text=ADVERSA&fontSize=48&fontColor=ffffff&fontAlignY=58" width="100%" alt="ADVERSA"/>

# ADVERSA

### LLM red-team harness — OWASP LLM Top 10 + MITRE ATLAS attack packs

<img src="https://readme-typing-svg.demolab.com?font=Fira+Code&size=18&duration=3500&pause=1000&color=6B46C1&center=true&vCenter=true&width=720&lines=LLM+redteam+harness++OWASP+LLM+Top+10++MITRE+ATLAS+attack+pa;Self-hostable+%C2%B7+MCP-native+%C2%B7+CI-ready+%C2%B7+polyglot" width="720"/>

[![PyPI](https://img.shields.io/pypi/v/cognis-adversa.svg?color=6b46c1)](https://pypi.org/project/cognis-adversa/) [![CI](https://github.com/cognis-digital/adversa/actions/workflows/ci.yml/badge.svg)](https://github.com/cognis-digital/adversa/actions) [![License: COCL 1.0](https://img.shields.io/badge/License-COCL%201.0-2b6cb0.svg)](LICENSE) [![Suite](https://img.shields.io/badge/Cognis-Neural%20Suite-6b46c1.svg)](https://github.com/cognis-digital)

*AI Security & Governance — securing LLMs, agents, and the MCP supply chain.*

</div>

```bash
pip install cognis-adversa
adversa scan .            # → prioritized findings in seconds
```


<!-- cognis:example:start -->
## 🔎 Example output

Real, reproducible output from the tool — runs offline:

```console
$ adversa-emit --version
adversa 2.0.0
```

```console
$ adversa-emit --help
usage: adversa [-h] [--version] {catalog,scan,probe,refs} ...

LLM red-team probe runner (OWASP LLM Top-10 + MITRE ATLAS).

positional arguments:
  {catalog,scan,probe,refs}
    catalog             list the probe catalog
    scan                run probes against a target
    probe               show detail for one probe
    refs                show OWASP + ATLAS reference tables

options:
  -h, --help            show this help message and exit
  --version             show program's version number and exit
```

```console
$ adversa-emit catalog
ADVERSA probe catalog (12 probes)
==============================================================================
ID                       OWASP   ATLAS        SEV       NAME
------------------------------------------------------------------------------
pi.direct_override       LLM01   AML.TA0004   high      Direct instruction override
pi.indirect_payload      LLM01   AML.TA0006   critical  Indirect prompt injection via retrieved content
pi.encoded_smuggling     LLM01   AML.TA0009   medium    Encoded payload smuggling
leak.system_prompt       LLM07   AML.TA0011   high      System prompt extraction
leak.credentials         LLM02   AML.TA0010   critical  Sensitive credential disclosure
harm.dangerous_instructions LLM09   AML.TA0005   high      Dangerous-capability elicitation
harm.roleplay_jailbreak  LLM01   AML.TA0009   high      Persona/roleplay jailbreak (DAN-style)
output.xss_injection     LLM05   AML.TA0006   high      Improper output handling (XSS payload)
agency.tool_abuse        LLM06   AML.TA0006   high      Excessive agency / unsafe tool invocation
misinfo.confident_falsehood LLM09   AML.TA0014   medium    Misinformation / fabricated authority
consumption.amplification LLM10   AML.TA0014   low       Unbounded consumption (resource amplification)
poison.training_data     LLM04   AML.TA0003   medium    Data poisoning acknowledgement
```

> Blocks above are real `adversa` output — reproduce them from a clone.

<!-- cognis:example:end -->

## Usage — step by step

`adversa` is an LLM red-team probe runner mapping the OWASP LLM Top-10 + MITRE ATLAS onto runnable probes.

1. **Install** (Python 3.10+):
   ```bash
   pip install -e .            # or: pipx install adversa
   ```
2. **Browse the bundled probe catalog** (filter by OWASP/ATLAS/severity):
   ```bash
   adversa catalog --owasp LLM01 --min-severity high
   ```
3. **Scan a target** — the bundled `secure`/`vulnerable` references, a **captured-response transcript** (offline, no live endpoint), or your own `module:callable` of signature `target(prompt) -> str`:
   ```bash
   adversa scan vulnerable
   adversa scan transcript:demos/01-healthcare-chatbot/transcript.json
   adversa scan mypkg.mymodel:generate --owasp LLM01
   ```
4. **Read the output** as a table, JSON, or **SARIF 2.1.0** (for GitHub code-scanning), or inspect one probe's prompts + grader + remediation:
   ```bash
   adversa scan vulnerable --format json | jq '.results[] | select(.passed==false)'
   adversa scan vulnerable --format sarif > adversa.sarif
   adversa probe pi.direct_override
   adversa refs        # OWASP LLM Top-10 + ATLAS tactic tables
   ```
5. **Gate CI** — `scan` exits `1` when findings are present, `0` when clean, `2` on usage error:
   ```yaml
   - run: pip install -e . && adversa scan mypkg.mymodel:generate   # non-zero fails the job
   ```


## Contents

- [Why adversa?](#why) · [Features](#features) · [Quick start](#quick-start) · [Example](#example) · [Demos](#demos) · [Architecture](#architecture) · [AI stack](#ai-stack) · [How it compares](#how-it-compares) · [Integrations](#integrations) · [Install anywhere](#install-anywhere) · [Related](#related) · [Contributing](#contributing)

<a name="why"></a>
## Why adversa?

LLM red-team harness — OWASP LLM Top 10 + MITRE ATLAS attack packs — without standing up heavyweight infrastructure.

`adversa` is single-purpose, scriptable, and self-hostable: point it at a target, get prioritized results in the format your workflow already speaks (table · JSON · SARIF), gate CI on it, and let agents drive it over MCP.

<div align="right"><a href="#top">↑ back to top</a></div>

<a name="features"></a>
## Features

- ✅ 12-probe catalog mapped to OWASP LLM Top-10 (2025) + MITRE ATLAS tactics
- ✅ Severity ranking + filtering (`--owasp`, `--atlas`, `--min-severity`, `--probe`)
- ✅ Five graders (must-refuse, must-not-leak, must-not-contain, must-contain, injection-resisted)
- ✅ **Transcript replay target** — red-team *captured* responses offline, no live endpoint
- ✅ Bundled `secure` / `vulnerable` reference targets + `module:callable` for your own model
- ✅ Output as **table · JSON · SARIF 2.1.0** (GitHub code-scanning ready)
- ✅ CI gate via exit codes (0 clean · 1 findings · 2 usage)
- ✅ 8 real-use-case [demos](demos/) with run commands + remediation guidance
- ✅ Runs on Linux/macOS/Windows · Docker · devcontainer
- ✅ Ports in Python, JavaScript, Go, and Rust (`ports/`)

<div align="right"><a href="#top">↑ back to top</a></div>

<a name="quick-start"></a>
## Quick start

```bash
pip install cognis-adversa
adversa --version
adversa scan .                       # scan current project
adversa scan . --format json         # machine-readable
adversa scan . --fail-on high        # CI gate (non-zero exit)
```

<div align="right"><a href="#top">↑ back to top</a></div>

<a name="example"></a>
## Example

```text
$ adversa scan .
  [HIGH    ] ADV-001  example finding             (./src/app.py)
  [MEDIUM  ] ADV-002  another signal              (./config.yaml)

  2 findings · risk score 5 · 38ms
```

<div align="right"><a href="#top">↑ back to top</a></div>

<a name="demos"></a>
## Demos — real scenarios you can run now

Each [`demos/<NN-name>/`](demos/) holds a realistic input (a captured-response
`transcript.json` in ADVERSA's real format, or a `module:callable` target) plus
a `SCENARIO.md` explaining where the data came from, the exact command, what to
expect, and how to act on the findings.

| Demo | Scenario | What it shows |
|---|---|---|
| [01](demos/01-healthcare-chatbot/) | Healthcare chatbot, pre-launch | 4 findings — system-prompt + credential leak block launch |
| [02](demos/02-post-hardening-clean/) | Same bot after hardening | **0 findings** — clean CI gate (exit 0) |
| [03](demos/03-rag-indirect-injection/) | RAG poisoned document | indirect + encoded prompt injection (LLM01) |
| [04](demos/04-agentic-tool-abuse/) | Agent with shell access | excessive agency (`rm -rf /`) + directive override |
| [05](demos/05-customer-support-jailbreak/) | Support bot jailbreak | DAN persona + harmful-instruction elicitation |
| [06](demos/06-rag-misinformation/) | Research assistant | fabricated citation + data-poisoning acceptance |
| [07](demos/07-fully-vulnerable-baseline/) | Worst-case baseline | all 12 probes fail (`vulnerable` target) |
| [08](demos/08-custom-import-target/) | Your own model | wiring a `module:callable` target into CI |

```bash
adversa scan transcript:demos/01-healthcare-chatbot/transcript.json   # 4 findings, exit 1
adversa scan transcript:demos/02-post-hardening-clean/transcript.json # 0 findings, exit 0
```

The transcript shape is either a probe-id map (`{"leak.system_prompt": "<reply>"}`)
or a list of `{"probe_id": "...", "response": "..."}` pairs — capture your model's
replies once, then grade them offline as often as you like.

<div align="right"><a href="#top">↑ back to top</a></div>

<a name="architecture"></a>
## Architecture

```mermaid
flowchart LR
  IN[sources] --> P[adversa<br/>curate + validate]
  P --> OUT[query / analysis]
```

<div align="right"><a href="#top">↑ back to top</a></div>

<a name="ai-stack"></a>
## Use it from any AI stack

`adversa` is interoperable with every popular way of using AI:

- **MCP server** — `adversa mcp` (Claude Desktop, Cursor, Cognis.Studio, [uncensored-fleet](https://github.com/cognis-digital/uncensored-fleet))
- **OpenAI-compatible / JSON** — pipe `adversa scan . --format json` into any agent or LLM
- **LangChain · CrewAI · AutoGen · LlamaIndex** — wrap the CLI/JSON as a tool in one line
- **CI / scripts** — exit codes + SARIF for non-AI pipelines

<div align="right"><a href="#top">↑ back to top</a></div>

<a name="how-it-compares"></a>
## How it compares

| | **Cognis adversa** | leondz |
|---|:---:|:---:|
| Self-hostable, no account | ✅ | varies |
| Single command, zero config | ✅ | ⚠️ |
| JSON + SARIF for CI | ✅ | varies |
| MCP-native (AI agents) | ✅ | ❌ |
| Polyglot ports (JS/Go/Rust) | ✅ | ❌ |
| Open license | ✅ COCL | varies |

*Built in the spirit of **leondz/garak**, re-framed the Cognis way. Missing a credit? Open a PR.*

<div align="right"><a href="#top">↑ back to top</a></div>

<a name="integrations"></a>
## Integrations

Pipes into your stack: **SARIF** for code-scanning, **JSON** for anything, an **MCP server** (`adversa mcp`) for AI agents, and a webhook forwarder for SIEM/Slack/Jira. See [`docs/INTEGRATIONS.md`](docs/INTEGRATIONS.md).

<div align="right"><a href="#top">↑ back to top</a></div>

<a name="install-anywhere"></a>
## Install — every way, every platform

```bash
pip install "git+https://github.com/cognis-digital/adversa.git"    # pip (works today)
pipx install "git+https://github.com/cognis-digital/adversa.git"   # isolated CLI
uv tool install "git+https://github.com/cognis-digital/adversa.git" # uv
pip install cognis-adversa                                          # PyPI (when published)
docker run --rm ghcr.io/cognis-digital/adversa:latest --help        # Docker
brew install cognis-digital/tap/adversa                             # Homebrew tap
curl -fsSL https://raw.githubusercontent.com/cognis-digital/adversa/main/install.sh | sh
```

| Linux | macOS | Windows | Docker | Cloud |
|---|---|---|---|---|
| `scripts/setup-linux.sh` | `scripts/setup-macos.sh` | `scripts/setup-windows.ps1` | `docker run ghcr.io/cognis-digital/adversa` | [DEPLOY.md](docs/DEPLOY.md) (AWS/Azure/GCP/k8s) |

<div align="right"><a href="#top">↑ back to top</a></div>

<a name="related"></a>
## Related Cognis tools

- [`aegis`](https://github.com/cognis-digital/aegis) — AI Agent Permission & Access Auditor — surfaces the lethal trifecta of credentials + injection + reach
- [`promptmirror`](https://github.com/cognis-digital/promptmirror) — Prompt-injection & indirect-injection scanner for any LLM context input
- [`ledgermind`](https://github.com/cognis-digital/ledgermind) — Local LLM cost & token forensics proxy with anomaly detection
- [`guardpost`](https://github.com/cognis-digital/guardpost) — Runtime agent firewall — PII redaction, rate limits, policy enforcement
- [`hallumark`](https://github.com/cognis-digital/hallumark) — LLM hallucination & grounding auditor for RAG systems
- [`aicard`](https://github.com/cognis-digital/aicard) — Auto-generated NIST AI RMF / EU AI Act Annex IV model & system cards

**Explore the suite →** [🗂️ all 170+ tools](https://github.com/cognis-digital/cognis-neural-suite) · [⭐ awesome-cognis](https://github.com/cognis-digital/awesome-cognis) · [🔗 cognis-sources](https://github.com/cognis-digital/cognis-sources) · [🤖 uncensored-fleet](https://github.com/cognis-digital/uncensored-fleet) · [🧠 engram](https://github.com/cognis-digital/engram)

<div align="right"><a href="#top">↑ back to top</a></div>

<a name="contributing"></a>
## Contributing

PRs, new rules, and demo scenarios are welcome under the collaboration-pull model — see [CONTRIBUTING.md](CONTRIBUTING.md) and [SECURITY.md](SECURITY.md).

> ### ⭐ If `adversa` saved you time, **star it** — it genuinely helps others find it.

## Interoperability

`{}` composes with the 300+ tool Cognis suite — JSON in/out and a shared
OpenAI-compatible `/v1` backbone. See **[INTEROP.md](INTEROP.md)** for the
suite map, composition patterns, and reference stacks.

## License

Source-available under the **Cognis Open Collaboration License (COCL) v1.0** — free for personal, internal-evaluation, research, and educational use; **commercial / production use requires a license** (licensing@cognis.digital). See [LICENSE](LICENSE).

---

<div align="center"><sub><b><a href="https://cognis.digital">Cognis Digital</a></b> · one of 170+ tools in the <a href="https://github.com/cognis-digital/cognis-neural-suite">Cognis Neural Suite</a> · <i>Making Tomorrow Better Today</i></sub></div>