oscal agent guardrails.stub

---

id: oscal-agent-guardrails name: OSCAL Agent Guardrails repo_path: ~/Documents/sw30labs/repos/oscal-agent-guardrails remote_url: https://github.com/sw30labs/oscal-agent-guardrails.git primary_language: Python framework: LangGraph ingested: 2026-04-11 last_commit_date: 2026-04-11 category: oscal-compliance stacks: [oscal, agentic, langgraph, compliance]

OSCAL Agent Guardrails

Purpose

Uses OSCAL profiles as a policy brain to guardrail LLM agents at runtime. The system implements a three-stage workflow where a planner LLM proposes tool calls, a policy enforcer consults the OSCAL profile to allow/deny/request approval for each tool, and a responder LLM explains the outcome to the user.

Architecture

LangGraph sequential workflow with 3 nodes:

1. Planner LLM — proposes tool calls based on user request
2. Policy Enforcer — OSCAL engine that checks proposed tool against loaded JSON profile
3. Responder LLM — explains allow/deny/approval_needed decision

Policy loaded from JSON profile. Implements a whitelist pattern — unknown tools are denied by default.

Dependencies

• langgraph
• langchain
• langchain-openai
• pydantic
• python-dotenv

CI/CD

GitHub Actions ci.yml

Tags

python langgraph langchain oscal nist security compliance

Cross-references

• oscal-digital-twin-playground
• oscal-zero-trust-lens
• oscal-agent-lab
• oscal-cac-playgd