Multi Agent Memory

多 agent 共享记忆与项目协作架构。支持项目状态隔离、知识库共享、跨项目搜索、版本控制、里程碑跟踪、周报和交接文档。适用于多个 agent 协作开发多个项目的场景。

This skill appears to be a local project/knowledge-file manager (no network exfiltration in the scripts), but it will read and write many files under /root/.openclaw and ~/workspace-<agent> even though the registry metadata does not declare those config paths. Before installing or enabling it: 1) Review the templates directory and scripts (they are plain shell) to ensure no sensitive data will be overwritten. 2) Run the skill in an isolated or non-root environment first (do not give it access to your real /root or production data). 3) If you expect to keep secrets elsewhere, confirm the skill won't read those paths. 4) Note the SKILL.md has small inaccuracies (uses 'read' where 'cat' would be expected) — test the scripts manually. 5) If you need tighter control, ask the author to declare the required config paths in the manifest and to limit file access to a configurable working directory.

Type: OpenClaw Skill Name: multi-agent-memory Version: 0.1.0 The skill bundle provides a structured framework for multi-agent collaboration and project memory management. It uses shell scripts (e.g., init-project.sh, daily-check.sh) and markdown templates to automate project initialization, status tracking, and reporting within the /root/.openclaw/ directory. The logic is consistent with its stated purpose, and there is no evidence of data exfiltration, unauthorized network access, or malicious prompt injection.