Skip to content

Releases: ARPAHLS/aura

v0.3.4 — Reference ToolHost coat, CLI depth, and onboarding

Choose a tag to compare

@rosspeili rosspeili released this 26 Aug 13:18

AURA Harness — a runtime coat for agent loops: audit, policy, and compliance export.

v0.3.4 packages everything merged since v0.3.3. The headline is the reference ToolHost epic (#12): a host-agnostic membrane coat you wrap around real skill runtimes without replacing the loop inside. Manifest rules merge at bind, Monitor and Break observer presets watch the spine, ingress bind context lands on registration events, OTel exports promoted principal attributes, and the sequencer can skip steps conditionally with when.

On top of that coat, this release adds CLI and docs depth — hash-chain verification, audit report display, agent and config management, flat runnable examples, an integrations index, and a first-run onboarding guide.

Experimental (alpha). API may change before 1.0.


Install

pip install aura-harness==0.3.4

Optional Skillware integration:

pip install "aura-harness[skillware]==0.3.4"

From source:

git clone https://github.com/ARPAHLS/aura.git
cd aura
git checkout v0.3.4
pip install -e ".[dev]"

Reference ToolHost coat (#12)

  • ToolHost protocol and SkillwareHost reference adapter — wrap execute()-style runtimes through egress policy
  • Skill manifest merge at bind — session rules enriched from skill manifest at registration
  • Monitor observer preset — after-call analytics on the spine
  • Break observer preset — alerts on repeated tool intents
  • Ingress bind enrichment — host bind context on skill registration events
  • OTel promoted attributes — agent ref, policy version, principal, and skill id on spans
  • Sequencer when — conditional step skip with skipped-step events on the spine
  • Capstone guide — end-to-end mock → live → Ollama path in one doc
  • Examples 05–08 — skill tour, sequencer chain, observer presets, emit-only loose coat
  • Integrations — Skillware adapter index, registry loader, mock/live reference scripts, cloud body loops (OpenAI, Anthropic, Gemini)
  • [integrations] extra — optional deps for skillware, ollama, and cloud providers

CLI & audit receipts

  • Hash-chain verification for exported JSONL sessions
  • Human-readable audit reports, JSON output for CI
  • Agent profile editing, merged config display, persisted project and storage paths
  • Interactive splash, grouped help, splash polish on Windows Terminal
aura verify chain path/to/session.jsonl
aura report show
aura report show --json

Docs & examples

  • Onboarding guide — posture → identity → body → approve → receipt
  • Session close and export — audit reports, export commands, receipt review in using-aura.md
  • Flat examples layout (#21) — runnable scripts at repo root with a single README index
  • Docs sync — INDEX, ROADMAP, integration guides aligned with epic closure

CI & quality

  • Reusable test workflow (#17) — PR CI and PyPI publish share the same lint/test gate
  • Python 3.10–3.13 matrix on Ubuntu (#10)
  • Core test gaps closed (#4) — tamper detection, compare fields, constraint matrix
  • Warn-only pip-audit in CI (#11)

Changed

  • Example 06 compress step skips when scan reports unsafe (sequencer when)
  • Agent registry keeps ref/alias maps consistent when agent ref changes
  • Optional persisted project directory in global config

Links

Full diff: v0.3.3...v0.3.4


Contributors

@rosspeili @DebayanSen96 @evgenyponomarev @bladedancer-chronicle3 @frobel0520

v0.3.3 — PR CI, test suite, and publish workflow fix

Choose a tag to compare

@rosspeili rosspeili released this 21 Aug 05:25

Summary

Quality and docs release: required PR checks, expanded test coverage, stronger comparison doc, and a fix for duplicate PyPI publishes.

Install

pip install aura-harness==0.3.3

Added

  • PR CI.github/workflows/ci.yml runs lint-test (black, flake8, pytest) on every PR and push to main.
  • Test suite — shared tests/conftest.py; test_cli.py, test_core_gaps.py, test_examples_smoke.py (49 tests, all example scripts smoke-tested).
  • Coverage reportpytest --cov=aura in CI (report only, no gate).

Changed

  • docs/comparison.md — membrane vs prompt-harness positioning; session receipt language.
  • Testing / contributing / publishing docs — CI parity and publish trigger guidance.

Fixed

  • PyPI publish — removed duplicate tag trigger; skip-existing: true for safe re-runs.

Links

Full diff: v0.3.2...v0.3.3

v0.3.2 — Deployments sidebar and README badge cleanup

Choose a tag to compare

@rosspeili rosspeili released this 20 Aug 16:24

Patch release.

Changed

  • Publish workflow registers GitHub pypi environment with PyPI URL (Deployments in repo sidebar).

Fixed

  • README badges: single row — Version, DOI, License, Powered by.
pip install -U aura-harness

v0.3.1 — PyPI README and badge fixes

Choose a tag to compare

@rosspeili rosspeili released this 20 Aug 16:05

Patch release — no API changes.

Fixed

  • README splash image uses raw GitHub URL so the logo renders on PyPI
  • Version badge uses pastel styling (matches other badges)
  • Two-row badge layout with DOI badge included on PyPI project page

Install

pip install -U aura-harness

Full diff: v0.3.0...v0.3.1

v0.3.0 — Layered identity, audit report, and enterprise export

Choose a tag to compare

@rosspeili rosspeili released this 20 Aug 15:46

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

v0.3.0 adds layered identity, a structured audit report with hash-chain verification, OTel export, and run comparison — on top of the v0.2 membrane, sequencer, Skillware host, and observers.

Experimental (alpha). API may change before 1.0.


Install

pip install aura-harness

Optional Skillware integration:

pip install "aura-harness[skillware]"

From source:

git clone https://github.com/ARPAHLS/aura.git
cd aura
pip install -e ".[dev]"

Added in v0.3.0

Identity

  • agent_ref (tenant/slug) for humans and CI
  • ULID internal aura_id, optional user-supplied aura_id
  • ids.tenant and policy_version on profile and spine trailer

Audit

  • AuditReport on session close — verdict, scorecard, findings, recommendations
  • Hash chain on every spine event (prev_hash / content_hash)
  • approve(request_id, principal=...) records approver on spine

Export & CLI

  • aura export-otel and .otel.jsonl on session close
  • aura compare <session_a> <session_b>

Developer experience

  • black + flake8 in [dev]
  • docs/TESTING.md
  • Contributor guide, agent workflow, and inclusive Code of Conduct

Changed in v0.3.0

  • Registry defaults to ULID ids; resolve() by ref, name, or id; legacy AURA-000n profiles still load
  • README refresh — minimal layout, linked badges, PyPI version badge
  • Private Manifesto repo references removed from public docs

Shipped in v0.2.0 (included in this line)

  • Membrane — ingress context, egress guarded_tool_call
  • Sequencer — prescriptive step pipelines with gates and retries
  • Skillware host — wrap skill execute() through egress
  • Observers — parallel dispatch on spine events
  • Example 04: research → draft → approve → notify

Shipped in v0.1.0 (kernel)

  • Agent registry, session lifecycle, audit spine, constraint engine
  • Conformance summary, JSONL export, CLI (aura agent, aura run, aura export)
  • Examples 01–03

Quick start

aura agent create --name demo
aura run examples/01-minimal-loop/run.py

Compare two sessions:

aura compare .aura/sessions/<session_a> .aura/sessions/<session_b>

Documentation


Full diff

v0.2.0...v0.3.0


Support: issues · systems@arpacorp.net