Skip to content

Releases: 412984588/openclaw-alignment

OpenClaw Alignment 2.0.0

Choose a tag to compare

@412984588 412984588 released this 08 Mar 07:13

Highlights

  • Finalizes the 2.0.0 policy-only public surface.
  • Canonical local storage is .openclaw_memory/policy/ with rules.json, playbooks.json, and policy_events.jsonl.
  • Supervision is centered on openclaw-align policy ....
  • Stable public API includes ConfirmationAPI, create_api, Rule, Playbook, PolicyEvent, PolicyStore, MarkdownToPolicyConverter, and PolicyToMarkdownExporter.

Validation

  • python -m build
  • python -m twine check dist/*
  • python -m pytest tests/ -v --tb=short (169 passed)
  • python scripts/check_docs_consistency.py
  • python -m ruff check lib tests scripts
  • python -m openclaw_align policy status
  • openclaw-align policy status

v1.0.1

Choose a tag to compare

@412984588 412984588 released this 01 Mar 22:59

v1.0.1

Fixed

  • Legacy bilingual GEP compatibility: Restore backward compatibility for legacy GEP data containing Chinese aliases
    • Fixed gep_to_md.py to match bilingual (English + Chinese) alias keys
    • Fixed md_to_gep.py to extract bilingual section mappings during migration
    • Preserved SOUL capsule semantics during Markdown-to-GEP conversion

Added

  • Compatibility regression tests: Added tests/test_gep_compatibility.py
    • Covers legacy Chinese data export/import roundtrip
    • Validates bilingual alias handling

Changed

  • Release hardening: Strengthened quality gates
    • Added ConfirmationAPI bootstrap validation
    • Ensured all 132 tests pass before release
    • Hardened documentation consistency checks

Validation

  • ✅ pytest full suite: 132 passed
  • ✅ mypy strict check: pass
  • ✅ ruff check: pass
  • ✅ docs consistency check: pass
  • ✅ build + twine metadata check: pass

Risk

Low-medium. Behavior changes are compatibility-oriented and guarded by regression tests.

v1.0.0: Production-ready OpenClaw Alignment

Choose a tag to compare

@412984588 412984588 released this 01 Mar 08:57

🎉 OpenClaw Alignment 1.0.0

首个生产就绪版本,提供强化学习驱动的工作流对齐引擎。

✨ 主要特性

  • Actor-Critic RL 核心引擎:完整的策略梯度学习和价值评估
  • 四维度奖励系统:客观指标、行为信号、显性反馈、模式偏好
  • 可选 Phase3 模块:分布式训练、自动调参、监控、性能优化
  • 契约防漂移:状态/动作维度与文档一致性自动检查
  • 跨平台支持:Windows、macOS、Linux

📦 安装

# 核心版本
pip install openclaw-alignment

# 完整版本(包含 Phase3 可选依赖)
pip install "openclaw-alignment[phase3]"

🔧 核心修复

  • 环境输入规范化(task_type 回退、time_of_day clamp)
  • 配置持久化保真(保留结构、自动创建目录)
  • 推理确定性(no-explore 模式改为贪心选择)
  • CLI 可靠性(修复 --help 崩溃)
  • Phase3 运行时优雅降级(Redis/Celery 不可用时自动回退)

📊 质量保证

  • 测试覆盖:80 个测试用例,100% 通过率
  • 性能:reward ~253k ops/s, agent ~55k ops/s
  • 治理:完整的 SECURITY、CONTRIBUTING、RELEASING 文档

📋 完整变更日志

参见 CHANGELOG.md

🙏 致谢

感谢所有贡献者和早期采用者的反馈与支持!


发布日期:2026-03-01
Python 支持:3.10, 3.11, 3.12, 3.13
许可证:MIT

What's Changed

  • 稳定性优化:多头策略、奖励修正与回放修复 by @412984588 in #1

Full Changelog: https://github.com/412984588/openclaw-alignment/commits/v1.0.0