Skip to content

Releases

aster 1.54.0 — runtime safety + refreshed introduction

Choose a tag to compare

@Hawse1337 Hawse1337 released this 20 Nov 14:10

aster 1.54.0 — runtime-safety boundary + refreshed introduction

This release refreshes the packaging/introduction and documents how the design bench is actually
executed: real AEDT runs go through the hardened guard, not loose PyAEDT calls.

What's new

  • Runtime safety section in the README and a new docs page
    user_doc/safety.md:
    managed-only AEDT sessions, unmanaged-session refusal, arbitrary-code default-off, run-token
    scope, launch/release audit, and the fail-closed client rule for DeepSeek Harness and Codex.
  • Refreshed landing page (hawse1337.github.io/aster) with
    the three delivery forms and a runtime-safety column; AEDT-integration page now states the guard
    version and the boundary.
  • README updated: three forms (library/CLI, agent skill, docs+releases), runtime-safety
    construction, real-study prerequisites against guard v0.3.2+.
  • Version bumps: pyproject.toml, aster/version.py, CITATION.cff, release references.
  • No library API change; offline suite unchanged: 782 passed, 80 subtests.

Assets

  • aster-1.54.0-py3-none-any.whl
  • aster-1.54.0.tar.gz

Related guard release

Real AEDT execution requires ansys-aedt-mcp
v0.3.2+, whose release is also mirrored on the guard release address
pyaedt-mcp-hardening-kit.

MIT licensed.


Same-version rebuild (2026-11-21)

Assets rebuilt from main at commit 513617b (still version/tag 1.54.0). Includes the scoped guard arbitrary-code opt-in in aster/bridge.py and the corrected runtime-safety documentation; 782 tests passing.

Bridge connect addendum (2026-11-21)

The trusted-bridge opt-in is now set before the initial connect call as well, so the broker worker spawned by connect inherits it; execute_code then runs modelling snippets without a persistent process-wide opt-in. Assets refreshed again under the same version/tag.

aster 1.53.0 — library · skill · docs

Choose a tag to compare

@Hawse1337 Hawse1337 released this 20 Nov 12:45

aster 1.53.0 — three-layer packaging

The bench is now shipped in the three forms promised in the README.

1. Python library + CLI

  • wheel/sdist attached below; extras: [optimizer,surrogate,plot]
  • CLI additions:
    • aster doctor — checks AEDT, guard checkout, guard venv, PyAEDT, reference project, campaign root
    • aster skill install|path|show — install the bundled agent skill
    • existing aster version, aster cases, aster dry-demo

2. Agent skill

  • canonical skill in skills/aster/: SKILL.md + references/ (API map, AEDT contract, evidence rules) + scripts/dry_demo.py
  • mirrored into the wheel as aster/skills/aster/, so aster skill install works from a pip install

3. Docs site + releases

Verification

  • Offline suite: 782 passed, 80 subtests (skill-bundle consistency tests included)
  • mkdocs build --strict passes locally; the Pages workflow deploys the site
  • aster skill install verified from both the source checkout and the built wheel
  • CI: Windows / CPython 3.12

MIT licensed. The web GUI is still under validation and is not part of this release.

aster 1.52.0

Choose a tag to compare

@Hawse1337 Hawse1337 released this 20 Nov 12:39

aster 1.52.0

An antenna design bench - evidence first. This is the renamed release of the project previously
published as antopt (1.51.0). The repository URL, Python import, CLI and environment prefix are now
aster / ASTER_*; the old antopt URL redirects to this repository.

What changed

  • Project renamed antopt ? aster; repository: https://github.com/Hawse1337/aster
  • Python package and import: aster (python -m aster, console script aster)
  • Environment prefix: ASTER_AEDT_EXE, ASTER_GUARD_REPO, ASTER_VENV_PYTHON, .
  • Docs, banner/loop illustrations and the agent skill updated to the new identity
  • First release of the redesign: banner, five-station loop diagram, bilingual README, skill install guide

What it is

  • Design-side library: parameter spaces, objectives/constraints, campaigns, surrogates, Pareto tools
  • Measured-run engine with artifacts kept next to every run (result.s1p, far-field JSON, native PNGs)
  • Explicit acceptance rules for S11 band and ring-pattern metrics; re-score history without re-solving
  • Real solves delegated to ansys-aedt-mcp (PyAEDT 1.10.0, AEDT 2026 R1)

Two delivery forms

  1. Python library - wheel/sdist attached here; pip install aster-1.52.0-py3-none-any.whl[optimizer,surrogate]
  2. Agent skill - SKILL.md in the repository; copy it into your agent's skill directory:
New-Item -ItemType Directory -Force "$env:USERPROFILE\.dsh\skills\aster" | Out-Null
Copy-Item SKILL.md "$env:USERPROFILE\.dsh\skills\aster\SKILL.md"

Verification

  • Offline suite: 778 passed, 80 subtests (python -m pytest tests -q), no AEDT needed
  • CLI: python -m aster version ? 1.52.0; python -m aster dry-demo --samples 64
  • CI: Windows / CPython 3.12

The pre-rename release 1.51.0 (package name antopt) stays available for history.

MIT licensed.

antopt 1.51.0

Choose a tag to compare

@Hawse1337 Hawse1337 released this 20 Nov 10:35

antopt 1.51.0

Physics-guided antenna design and optimisation library, with measured runs as evidence.

This is the library release. The web GUI/desktop app (HTTP server, static assets, launcher,
installer) is still under validation and is intentionally not published.

Highlights

  • Core contracts: ParamSpec/ParamSpace, Objective/Constraint, Evaluation, dominance and
    Pareto archive.
  • Campaign layer with a measured-run engine (antopt.runtime), campaign recipes and a run archive.
  • Acceptance rules kept in code: S11 band checks from Touchstone (band_acceptance), ring/pattern
    checks (pattern_acceptance) and a combined verdict (acceptance).
  • Surrogates: Gaussian-process and multi-fidelity models with explicit validation state.
  • AEDT integration is delegated to ansys-aedt-mcp
    (PyAEDT 1.10.0, AEDT 2026 R1); every real run keeps its S11, far-field and native-report artifacts.
  • Agent packaging: repository-level SKILL.md for agent-driven optimisation workflows.

Verification

  • Offline suite: 778 passed, 80 subtests, 1 skipped (python -m pytest tests -q), no AEDT needed.
  • python -m antopt version ? 1.51.0; python -m antopt cases; python -m antopt dry-demo --samples 64.
  • Wheel and sdist are attached below.

Install

py -3.10 -m venv .venv
.\.venv\Scripts\python.exe -m pip install -U pip
.\.venv\Scripts\python.exe -m pip install antopt-1.51.0-py3-none-any.whl[optimizer,surrogate]
python -m antopt version

Documentation

MIT licensed.