AURA Harness - A runtime coat for agent loops — audit, policy, and compliance export.
Overview · How it works · Architecture · Quick start · Documentation
AURA Harness wraps whatever runs your agent loop — a Python script, Skillware, LangGraph, or your own host. It does not replace the loop. It sits around it, records what happened, enforces your rules at tool boundaries, and exports a session you can ship to logs or compliance.
| For | Teams that need provenance, policy, and repeatable pipelines around agents |
| Not | A model runtime, orchestrator, or skill framework |
| Pairs with | Skillware (tools), any LLM host, your existing loop |
flowchart LR
Host[Host] --> Aura[AURA]
Aura --> Audit[Audit Trail]
Your host runs the loop. AURA attaches — policy at egress, causal logging. On close, the audit trail becomes session export (JSONL, summary, audit report).
→ using-aura.md · skillware-integration.md
Optional inputs feed the body; AURA wraps it; audit and export follow.
flowchart LR
ID["Identity"] -.-> BODY["Body / Runtime"]
BRAIN["Brain"] -.-> BODY
MEM["Memory"] -.-> BODY
TOOLS["Tools"] -.-> BODY
CONST["Constitution"] -.-> BODY
BODY --> AURA["AURA Harness"]
AURA --> TRAIL["Audit Trail"]
TRAIL --> EXPORT["Session Export"]
TRAIL -.-> OBS["Observers"]
→ architecture.md · stack-position.md
Requires Python 3.10+.
Install from PyPI:
pip install aura-harness
# optional Skillware host extra:
pip install "aura-harness[skillware]"Develop from source:
git clone https://github.com/ARPAHLS/aura.git
cd aura
py -3.13 -m venv .venv && .venv\Scripts\activate # Windows
pip install -e ".[dev]"
pytestfrom aura import agent, configure
configure()
ag = agent("acme/research-bot", policy_version="1")
with ag.session() as run:
run.emit("turn.start", {"input": "hello"})
run.emit("turn.end", {"tokens": 120})
print(run.exports)CLI: aura agent create, aura run, aura export, aura compare, aura export-otel.
→ getting-started.md · examples/
| Topic | Links |
|---|---|
| Start | getting-started.md · concepts.md · using-aura.md |
| Integration | skillware-integration.md · sequencer.md |
| Identity & audit | trust-paths.md · outputs.md |
| Compare & position | comparison.md · ROADMAP.md |
| Contribute | CONTRIBUTING.md · Agent workflow · TESTING.md · PUBLISHING.md · CHANGELOG.md |
Notes
- Experimental (alpha) — API may change between minor releases until 1.0.
- Wrap, don't replace — your host owns the loop; AURA records and gates at boundaries you wire through
emit()orSkillwareHost. - Security — loaded skills and scripts run in your process; see SECURITY.md.
- Citation — CITATION.cff · DOI 10.5281/zenodo.22031863 (Zenodo concept DOI)
- Support — issues · systems@arpacorp.net
