github alibaba/hiclaw v1.0.8

one day ago

Highlights

What's New

  • OpenClaw Upgrade to v2026.3.8 — Upgraded the bundled OpenClaw to v2026.3.8, widely regarded as the most stable release by the community. This brings improved reliability and performance to the underlying agent runtime.

  • Alibaba Cloud Native Deployment — Added Alibaba Cloud native deployment support with a unified cloud/local abstraction layer, including CoPaw worker support for cloud deployment. Manager Dockerfile.aliyun now sets ENV HOME=/root/manager-workspace so the agent writes workspace files to the correct directory.

  • Env Var Placeholder Rendering — SKILL.md and AGENTS.md now render ${VAR} placeholders at startup via envsubst, so AI agents read plain text instead of raw environment variable references.

  • Unified AI Gateway URL — Refactored all scripts to use a single HICLAW_AI_GATEWAY_SERVER env var in hiclaw-env.sh, removing duplicated cloud/local branching from create-worker, generate-worker-config, and model-switch scripts.

  • Control UI Token Security — Control UI now uses URL hash redirect for token injection, and Manager generates a distinct hooks.token on first boot with proper base64 encoding (-w 0) to prevent JSON parse errors.

  • Worker Import System — New hiclaw-import.sh / hiclaw-import.ps1 scripts and migration skill allow importing Workers from external sources, with automated analysis and zip generation for seamless migration.

  • Agent Docs Progressive Disclosure — Restructured SKILL.md and AGENTS.md into a references/ directory structure for progressive disclosure, reducing token consumption. Added GitHub bug report template.

  • Debug Log Export Tool — New export-debug-log.py helps diagnose why an Agent's behavior doesn't meet expectations. It exports Matrix messages and agent session logs (OpenClaw / CoPaw) in one command, letting you trace the full conversation context, tool calls, and decision flow that led to an unexpected outcome. Supports time range (--range 1h/1d), container/room filtering, and built-in PII auto-redaction for safe sharing in bug reports.

  • Install Step-Back Navigation — Install scripts (sh/ps1) now support step-back navigation, allowing users to go back to previous steps during installation instead of restarting.

  • Docker Layer Optimization — Reordered Dockerfile layer ordering for Manager, Worker, and CoPaw images to maximize cache hits and speed up upgrade pulls.

  • Worker Security Hardening — Added mirror loop safeguard and security hardening for Workers, preventing agents from entering infinite mirror loops.

  • Higress mergeConsecutiveMessages — Enabled mergeConsecutiveMessages for all LLM providers in Higress gateway config, improving compatibility with models that reject consecutive same-role messages.

  • Infinite Task Loop Prevention — Improved error guidance in task-management and model-switch skills to prevent agents from entering infinite task loops.

Bug Fixes

  • Fixed GPT-5 model LLM connectivity tests — use max_completion_tokens instead of max_tokens for GPT-5 models; also removed max_tokens=1 from LLM connectivity tests across install scripts.

  • Fixed Worker name handling in create-worker.sh — normalize worker name to lowercase to match Tuwunel's username storage behavior, fixing invite failures when names contain uppercase letters; added validation to reject invalid characters before Matrix registration.

  • Fixed welcome message readiness in Aliyun deployment — use gateway health check instead of Matrix room member polling, with timeout increased to 300s.

  • Fixed skill removal cleanup — Manager's --remove-skill now deletes MinIO files and notifies worker; worker prunes stale skill dirs while preserving builtins.

新增功能

  • OpenClaw 升级至 v2026.3.8 — 将内置 OpenClaw 升级至 v2026.3.8,这是社区评价最稳定的新版本,带来了更好的可靠性和性能。

  • 阿里云原生部署 — 新增阿里云原生部署支持,提供统一的云/本地抽象层,包括 CoPaw Worker 的云端部署支持。Manager Dockerfile.aliyun 现在设置 ENV HOME=/root/manager-workspace,确保 Agent 将工作区文件写入正确目录。

  • 环境变量占位符渲染 — SKILL.md 和 AGENTS.md 现在在启动时通过 envsubst 渲染 ${VAR} 占位符,AI Agent 读取的是纯文本而非原始环境变量引用。

  • 统一 AI Gateway URL — 重构所有脚本使用统一的 HICLAW_AI_GATEWAY_SERVER 环境变量,移除 create-workergenerate-worker-configmodel-switch 脚本中重复的云/本地分支逻辑。

  • Control UI Token 安全增强 — Control UI 现在使用 URL hash 重定向注入 token,Manager 首次启动时生成独立的 hooks.token,并使用 base64 -w 0 编码防止 JSON 解析错误。

  • Worker 导入系统 — 新增 hiclaw-import.sh / hiclaw-import.ps1 脚本和 migration skill,支持从外部导入 Worker,提供自动化分析和 zip 生成实现无缝迁移。

  • Agent 文档渐进式披露 — 将 SKILL.md 和 AGENTS.md 重构为 references/ 目录结构实现渐进式披露,降低 token 消耗。新增 GitHub bug report 模板。

  • Debug Log 导出工具 — 新增 export-debug-log.py,帮助诊断 Agent 行为不达预期的原因。一条命令即可导出 Matrix 消息和 Agent 会话日志(OpenClaw / CoPaw),让你追溯完整的对话上下文、工具调用和决策流程,定位问题根因。支持时间范围过滤(--range 1h/1d)、容器/房间过滤,内置 PII 自动脱敏,可安全用于 bug report 分享。

  • 安装脚本回退导航 — 安装脚本(sh/ps1)现在支持回退导航,用户可以在安装过程中返回上一步而无需重新开始。

  • Docker 镜像层优化 — 重排 Manager、Worker 和 CoPaw 的 Dockerfile 层顺序,最大化缓存命中率,加速升级拉取。

  • Worker 安全加固 — 新增 mirror loop 防护和安全硬化,防止 Agent 进入无限镜像循环。

  • Higress 消息合并 — 为 Higress 网关所有 LLM provider 启用 mergeConsecutiveMessages,提升对拒绝连续同角色消息的模型的兼容性。

  • 无限任务循环防护 — 改进 task-management 和 model-switch skill 的错误引导,防止 Agent 陷入无限任务循环。

Bug 修复

  • 修复 GPT-5 模型 LLM 连通性测试 — 对 GPT-5 模型使用 max_completion_tokens 替代 max_tokens;同时移除安装脚本中 LLM 连通性测试的 max_tokens=1

  • 修复 create-worker.sh 中 Worker 名称处理 — 将 Worker 名称规范化为小写以匹配 Tuwunel 的用户名存储行为,修复名称含大写字母时的邀请失败问题;新增验证逻辑拒绝无效字符。

  • 修复阿里云部署中欢迎消息就绪检测 — 使用网关健康检查替代 Matrix 房间成员轮询,超时时间增加至 300 秒。

  • 修复技能移除清理 — Manager 的 --remove-skill 现在删除 MinIO 文件并通知 Worker;Worker 清理过时技能目录同时保留内置技能。

  • feat: upgrade openclaw to v2026.3.8 (7dea227)

  • feat(cloud): add Alibaba Cloud native deployment support with unified cloud/local abstraction layer (b98dcc8)

  • feat(cloud): add CoPaw worker support for cloud deployment (58a012f)

  • feat(agent): render env var placeholders in SKILL.md/AGENTS.md at startup via envsubst (7d18776)

  • refactor(manager): unify AI Gateway URL into HICLAW_AI_GATEWAY_SERVER in hiclaw-env.sh (4046019)

  • fix(manager): use distinct hooks.token on first boot (ac731a7)

  • fix(manager): use base64 -w 0 for hooks.token to prevent JSON parse error (d8ac9e8)

  • fix: use URL hash redirect for Control UI token injection (b3df8fe)

  • fix(manager): use max_completion_tokens for GPT-5 models in LLM connectivity tests (f3c6529)

  • fix: remove max_tokens=1 from LLM connectivity tests (4fc97bd)

  • fix(install): remove max_tokens=1 from LLM connectivity test in ps1 (7ff5182)

  • fix(manager): normalize worker name to lowercase in create-worker.sh (0bd2f58)

  • fix(manager): add worker name validation and set HOME in Dockerfile.aliyun (d800790)

  • fix(manager): use gateway health check for welcome message readiness in Aliyun deployment (ca83687)

  • fix(sync): clean up removed skills from MinIO, worker local, and active_skills (67ddcb6)

  • feat: add Worker import system and migration skill (f7c9e55)

  • feat: optimize agent docs, add debug log export and bug report template (4abd8cc)

  • feat(install): add step-back navigation support (9938330)

  • perf(docker): optimize layer ordering for faster upgrade pulls (c82c2af)

  • feat(agent): add mirror loop safeguard and security hardening for workers (ba60e95)

  • feat(higress): enable mergeConsecutiveMessages for all LLM providers (8c44c5b)

  • fix(manager): improve error guidance and prevent infinite task loops (586ff51)

Docker Images

Multi-architecture images (amd64 + arm64):

# Manager
docker pull higress-registry.cn-hangzhou.cr.aliyuncs.com/higress/hiclaw-manager:v1.0.8

# Worker
docker pull higress-registry.cn-hangzhou.cr.aliyuncs.com/higress/hiclaw-worker:v1.0.8

Quick Start

macOS / Linux

bash <(curl -fsSL https://raw.githubusercontent.com/alibaba/hiclaw/main/install/hiclaw-install.sh)

Windows (PowerShell 7+)

Set-ExecutionPolicy Bypass -Scope Process -Force; =New-Object Net.WebClient; .Encoding=[Text.Encoding]::UTF8; iex .DownloadString('https://higress.ai/hiclaw/install.ps1')

Documentation

For more details, see the installation guide.

Don't miss a new hiclaw release

NewReleases is sending notifications on new releases.