AURA Harness - A runtime coat for agent loops — audit, policy, and compliance export.

Run your agent on your env with your tools, AURA proves what happened and stops what shouldn't, without owning your loop or your credentials.

Overview · How it works · Architecture · Quick start · Documentation

---

Overview

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

---

How it works

flowchart LR
    Host[Host] --> Aura[AURA]
    Aura --> Audit[Audit Trail]

Loading

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

---

Architecture

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"]

Loading

→ architecture.md · stack-position.md

---

Quick start

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]"
pytest

from 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, aura verify chain.

→ getting-started.md · examples/

---

Documentation

Topic	Links
Start	getting-started.md · concepts.md · using-aura.md
Integration	reference-tool-host-capstone.md · guides/aura-on-skillware.md · 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() or SkillwareHost.
• 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

---

_{Developed and Maintained by ARPA HELLENIC LOGICAL SYSTEMS
Support: systems@arpacorp.net}