Metadata-Version: 2.4
Name: agentforge-guard-llamaguard
Version: 0.2.3
Summary: Llama Guard 3 classifier for AgentForge guardrails
Project-URL: Homepage, https://github.com/Scaffoldic/agentforge-py
Project-URL: Repository, https://github.com/Scaffoldic/agentforge-py
Project-URL: Documentation, https://github.com/Scaffoldic/agentforge-py
Project-URL: Changelog, https://github.com/Scaffoldic/agentforge-py/blob/main/CHANGELOG.md
Project-URL: Issues, https://github.com/Scaffoldic/agentforge-py/issues
Author: The AgentForge Authors
License-Expression: Apache-2.0
License-File: LICENSE
Keywords: agent,ai,guardrails,llama-guard
Classifier: Development Status :: 2 - Pre-Alpha
Classifier: Intended Audience :: Developers
Classifier: License :: OSI Approved :: Apache Software License
Classifier: Programming Language :: Python :: 3
Classifier: Programming Language :: Python :: 3.13
Classifier: Topic :: Security
Classifier: Topic :: Software Development :: Libraries :: Python Modules
Classifier: Typing :: Typed
Requires-Python: >=3.13
Requires-Dist: agentforge-core~=0.2.3
Description-Content-Type: text/markdown

# agentforge-guard-llamaguard

Meta [Llama Guard 3](https://huggingface.co/meta-llama/Llama-Guard-3-8B)
classifier for AgentForge guardrails (feat-018).

Adds the `llamaguard` validator to both `input` and `output`
sections of `modules.guardrails`. Unlike the other guard modules,
this one needs an `LLMClient` — Llama Guard runs as a chat model
producing one of `safe` / `unsafe S1..S14` for each turn:

```yaml
modules:
  guardrails:
    input:
      - llamaguard:
          model: "bedrock:meta.llama3-guard-3-8b-instruct-v1:0"
    output:
      - llamaguard:
          model: "bedrock:meta.llama3-guard-3-8b-instruct-v1:0"
```

```bash
agentforge add module guard-llamaguard
```

The validator constructs an `LLMClient` lazily from the model
string via the framework's resolver, so any provider that
implements `LLMClient` (Bedrock, local Ollama via a custom
provider, etc.) can host the guard model.
