← 返回 Skills 市场
zqh2333

Shared Memory Os

作者 Qihong · GitHub ↗ · v1.7.2 · MIT-0
cross-platform ⚠ suspicious
209
总下载
1
收藏
0
当前安装
16
版本数
在 OpenClaw 中安装
/install shared-memory-os
功能描述
Shared memory governance for multi-agent OpenClaw workspaces — with tiered memory, learnings capture, promotion review, lifecycle management, self-maintainin...
安全使用建议
This skill appears to implement the described shared-memory governance features, but review before installing: - Confirm the workspace path assumptions: many scripts use the hard-coded ROOT '/home/zqh2333/.openclaw/workspace'. If your environment differs, the scripts could read/write the wrong location or fail. Consider editing scripts to use a relative or configurable workspace root. - Ensure the required binaries exist: Node is required (node scripts are invoked) and the scripts call the 'openclaw' CLI and, in shell variants, 'jq'. The registry metadata did not declare these dependencies — install or declare them first. - Inspect the cron creation: the scripts will upsert scheduled jobs that run node commands with exec/read tooling. Confirm the exact cron messages/commands, schedules, and that jobs run in the intended 'sessionTarget' (isolated). If you do not want automated recurring runs, do not allow the cron upsert step. - Check for secrets and privacy: the skill explicitly tells agents to harvest learnings into .learnings/ and includes a private-secrets template advising to avoid including secrets in learnings. Still, verify that none of your secret files will be referenced/harvested automatically. - Run manually first: instead of enabling cron, run init-shared-memory-os.js and the reporting scripts manually in a controlled environment to verify behavior and outputs. Given the hard-coded paths and undeclared external tool assumptions, treat this skill as suspicious until you confirm and adapt it to your environment.
功能分析
Type: OpenClaw Skill Name: shared-memory-os Version: 1.7.2 The skill implements a 'Shared Memory OS' for workspace maintenance but exhibits high-risk behavior by instructing the agent to automatically establish persistence. Specifically, SKILL.md directs the agent to immediately create three recurring cron jobs with 'exec' and 'read' privileges. The scripts (ensure-shared-memory-crons.sh and .js) use the 'openclaw cron' command to schedule these tasks, which execute local Node.js scripts. Furthermore, the entire bundle relies on hardcoded absolute paths (/home/zqh2333/.openclaw/workspace), which is highly irregular for a portable skill and suggests targeting a specific environment. While no clear evidence of data exfiltration was found, the automated setup of persistent execution capabilities warrants a suspicious classification.
能力评估
Purpose & Capability
The skill claims to manage shared workspace memory and the shipped scripts implement that. However the bundle makes environment assumptions that are not declared: many scripts use absolute paths like '/home/zqh2333/.openclaw/workspace' and call external tools (openclaw CLI, node, jq) even though the registry metadata declares no required binaries or env vars. Those undeclared dependencies and hard-coded paths are disproportionate to a well-packaged workspace governance skill and may break or operate on an unexpected home directory.
Instruction Scope
SKILL.md instructs the agent to ensure recurring cron jobs and to run many scripts that read and write workspace files and reports. The cron job messages explicitly allow use of tools 'exec' and 'read' and ask agents to run and summarize outputs. The instructions advocate using a 'built-in cron' but the provided scripts shell out to the 'openclaw' CLI — a discrepancy. The runtime instructions and scripts will create, edit, and run scheduled jobs that execute arbitrary commands inside the workspace; this is within the skill's function but broad in scope and requires caution.
Install Mechanism
There is no remote install/download spec (instruction-only), which avoids fetching arbitrary remote executables. The package includes many local Node scripts that will be executed. Risk is moderate because the scripts assume Node and the openclaw CLI are present and will be run with filesystem write privileges. No external URLs or archives are used.
Credentials
The skill declares no required env vars or credentials, but scripts read environment variables at runtime (process.env.OPENCLAW_STATE_DIR, process.env.HOME) and expect the openclaw cron store at a specific location. They also implicitly require a writable workspace at a hard-coded absolute path. While no secrets are requested, the mismatch between declared env requirements (none) and actual environment use is concerning and could lead to running against unexpected directories.
Persistence & Privilege
The skill will create/update scheduled cron jobs (via the openclaw CLI) so it installs recurring automated behavior in the agent environment. always:false is set in metadata, but the created cron jobs are persistent and will run autonomously. That persistence is expected for a maintenance automation skill, but it increases the blast radius, so verify the exact commands and schedule before allowing them.
如何使用
  1. 确保已安装 OpenClaw(本地或 Docker 部署)
  2. 在对话框中输入安装命令:/install shared-memory-os
  3. 安装完成后,直接呼叫该 Skill 的名称或使用 /shared-memory-os 触发
  4. 根据 Skill 的参数说明提供必要输入,即可获得结构化输出
版本历史
v1.7.2
Prefer first-class OpenClaw cron tool for automatic recurring job creation in agent sessions; keep CLI bootstrap scripts as terminal fallbacks.
v1.7.1
Harden cron bootstrap/update scripts to use shell-based upsert flow for automatic shared-memory maintenance tasks.
v1.7.0
Add bootstrap and idempotent cron-install scripts for automatic shared-memory maintenance jobs; document first-run automation.
v1.6.0
Add collaborative governance upgrades: evidence/confidence outputs, validated-rule detection, low-value learning review, workflow optimization suggestions, conflict review, dashboard, maintenance-day guidance, and stronger cross-skill collaboration.
v1.5.0
Finalize the evolution system with one-page bilingual docs, maintenance runner, policy/workspace profiles, memory patch candidates, and audit reporting.
v1.4.0
Comprehensive upgrade: weekly review reports, promotion suggestions, health history, stale durable memory detection, merge suggestions, skill upgrade candidates, migration guide, conflict template, and bilingual one-page docs.
v1.3.1
Add a Chinese guide for Shared Memory OS and link it from the main skill documentation.
v1.3.0
Add health scoring, duplicate detection, promotion candidate detection, self-improving loop, and scheduled shared-memory maintenance.
v1.2.0
Add self-improving loop, memory health checks, learnings index rebuilding, and stronger shared-memory maintenance workflow.
v1.1.0
Comprehensive governance upgrade: layered memory rules, learnings harvesting, conflict resolution, review cadence, archive strategy, and reusable learnings template.
v1.0.5
Add heartbeat-to-front-memory promotion rules so summaries from maintenance threads can be recalled by all agents in main chat
v1.0.4
Add front-of-mind memory layers, active-thread continuity, success/default/feedback learning layers, and stronger governance navigation
v1.0.3
Add restricted secrets memory layer with explicit opt-in loading and governance rules
v1.0.2
Promote skill publishing governance into core shared-memory rules; add governed release-state handling
v1.0.1
Refined the external positioning of Shared Memory OS as a governance layer for multi-agent OpenClaw workspaces. Improved messaging around tiered memory, review cycles, conflict handling, and controlled evolution to better distinguish it from generic memory or long-term storage skills.
v1.0.0
Initial release: shared multi-agent memory operating system for OpenClaw workspaces, with tiered memory structure, heartbeat maintenance, weekly/monthly reviews, skill intake governance, conflict tracking, pattern harvesting, governance logs, dashboard, and sync rules.
元数据
Slug shared-memory-os
版本 1.7.2
许可证 MIT-0
累计安装 0
当前安装数 0
历史版本数 16
常见问题

Shared Memory Os 是什么?

Shared memory governance for multi-agent OpenClaw workspaces — with tiered memory, learnings capture, promotion review, lifecycle management, self-maintainin... 它是一个面向 Claude Code / OpenClaw 的 AI Agent Skill 插件,目前累计下载 209 次。

如何安装 Shared Memory Os?

在 OpenClaw 或 Claude Code 对话框中运行命令「/install shared-memory-os」即可一键安装,无需额外配置。

Shared Memory Os 是免费的吗?

是的,Shared Memory Os 完全免费,采用 MIT-0 许可证,可自由下载、安装和使用。

Shared Memory Os 支持哪些平台?

Shared Memory Os 跨平台运行,可在任意部署了 OpenClaw / Claude Code 的环境中使用(cross-platform)。

谁开发了 Shared Memory Os?

由 Qihong(@zqh2333)开发并维护,当前版本 v1.7.2。

💬 留言讨论