WorkClaw v0.5.7

WorkClaw Windows Release

• Highlights in v0.5.7:

  • 中文:
    • 新增了本地真实 agent 评测通道，可以用真实模型、真实 skill 和真实外部系统场景手动回放关键任务，并自动产出 pass / warn / fail 报告、trace、journal 和 stdout/stderr 证据。
    • 为真实评测补齐了匿名场景定义和本地私有配置映射，便于在不提交敏感 skill 路径与密钥的前提下，对 WorkClaw 的核心 skill 执行能力做回归验证。
    • 修复了真实评测中的结构化结果判定问题，现在可以正确识别像 summaries[].daily_facts / plan_facts / report_facts 这样的嵌套返回结构，减少误报失败。
    • 加强了本地模型配置的安全约束，api_key_env 现在要求填写环境变量名而不是明文 key，降低本地调试时把密钥写进配置和错误日志的风险。
    • 继续改进聊天运行时持久化，减少重复快照写入，同时保持会话恢复行为稳定。
  • English:
    • Added a local real-agent evaluation lane for replaying critical tasks with real models, real skills, and real external systems, with automatic pass / warn / fail reports plus trace, journal, and stdout/stderr artifacts.
    • Added anonymous scenario definitions and local-only capability mapping so WorkClaw can regression-test core skill execution without committing sensitive skill paths or secrets.
    • Fixed structured-result evaluation for real-agent runs so nested outputs such as summaries[].daily_facts / plan_facts / report_facts are interpreted correctly instead of being reported as false failures.
    • Hardened local model configuration safety by requiring api_key_env to be an environment-variable name rather than a literal API key, reducing the chance of leaking secrets into local config or error output.
    • Continued improving chat runtime persistence to reduce redundant snapshot writes while keeping session restore behavior stable.

• Recommended download: *-setup.exe for direct install.

• Enterprise deployment: *.msi for IT-managed installation and manual upgrades.

Installation Guide

1. Most users should install the setup.exe package.
2. Enterprise or managed devices can use the .msi package.

Verification Checklist

• Installer branding and Chinese setup wizard verified.
• Release tag matches desktop app version.