chernistry/bernstein

⚠️ Broken wheel — do not install. The PyPI wheel for v1.8.9 is missing 18 bernstein.core.* sub-packages, so bernstein --version crashes with ModuleNotFoundError: No module named 'bernstein.core.config'. Install v1.8.10 or later.

v1.8.9 — feature drop

OpenAI Agents SDK v2 adapter

A new CLI adapter wraps OpenAI's agents.Agent + Runner so agents built on the SDK become orchestratable inside a Bernstein plan.yaml, alongside Claude, Codex, and Gemini.

steps:
  - cli: openai_agents
    model: gpt-5-mini

Install: pip install 'bernstein[openai]'. MCP tools Bernstein manages bridge into the OpenAI runner; usage feeds the cost tracker.

Pluggable sandbox backends

Agent isolation is no longer git-worktree-only. A new SandboxBackend protocol ships with first-party worktree (default) and docker backends, plus optional e2b and modal extras. Plugin authors register more via bernstein.sandbox_backends.

stages:
  - sandbox:
      backend: docker
      options: { image: python:3.13-slim }

Cloud artifact storage sinks

.sdd/ state can now persist to S3, GCS, Azure Blob, or Cloudflare R2. A BufferedSink wrapper keeps the WAL fsync crash-safety invariant locally while mirroring to cloud asynchronously.

storage:
  sink: s3
  bucket: my-bernstein-runs

Extras: pip install 'bernstein[s3,gcs,azure,r2]'.

Progressive-disclosure skill packs

Role templates are now structured as templates/skills/<name>/SKILL.md + references/. A new load_skill MCP tool pulls skill bodies on demand — average system prompt drops ≥40% for the same role. Plugin skill packs register via bernstein.skill_sources.

bernstein skills list
bernstein skills show backend

Legacy templates/roles/*.md still resolve when no skill pack is present.

Documentation

Full docs refresh: README adapter count corrected, all four features surfaced with examples, comparison pages re-verified.

Install

pip install --upgrade 'bernstein>=1.8.10'

Full changelog: https://github.com/chernistry/bernstein/compare/v1.8.8...v1.8.9