Shared Memory OS

下方英文与中文在同一页面。在 ClawHub 中，直接阅读语言区块，无需依赖锚点跳转行为。

English

Shared Memory OS turns workspace memory into a governed system instead of a pile of notes.

Core capabilities

• - layered memory governance
• active learnings harvesting into .learnings/
• review cadence and archive strategy
• conflict resolution order
• health scoring and health history
• duplicate learnings detection with evidence/confidence
• promotion candidate detection with review state
• promotion suggestion and memory patch generation
• weekly review, audit report, and dashboard generation
• stale durable memory detection with lifecycle hints
• validated repeated rule detection
• low-value learning detection
• workflow optimization suggestions
• conflict logging and conflict review reporting
• learning merge suggestions
• skill upgrade candidate detection
• migration guidance and initialization scripts
• self-improving loop
• maintenance runner
• policy profiles and workspace profiles
• cross-skill collaboration governance
• multi-agent write and review guidance
• periodic maintenance-day guidance

References

• - references/governance-guide.md
• references/pattern-harvesting.md
• references/archive-strategy.md
• references/learnings-template.md
• references/health-check.md
• references/self-improving-loop.md
• references/health-score-model.md
• references/conflict-template.md
• references/migration-guide.md
• references/policy-profiles.md
• references/workspace-profiles.md
• references/cross-skill-collaboration.md
• references/multi-agent-governance.md
• references/maintenance-day.md
• references/zh-cn-guide.md

Scripts

• - scripts/bootstrap-shared-memory-os.js
• scripts/ensure-shared-memory-crons.js
• scripts/init-shared-memory-os.js
• scripts/migrate-notes-into-memory.js
• scripts/run-shared-memory-maintenance.js
• scripts/check-memory-health.js
• scripts/rebuild-learnings-index.js
• scripts/find-duplicate-learnings.js
• scripts/find-promotion-candidates.js
• scripts/find-stale-durable-memory.js
• scripts/find-validated-rules.js
• scripts/find-low-value-learnings.js
• scripts/build-workflow-optimization-suggestions.js
• scripts/log-memory-conflict.js
• scripts/build-conflict-review-report.js
• scripts/build-weekly-review.js
• scripts/record-health-snapshot.js
• scripts/build-promotion-suggestions.js
• scripts/build-learning-merge-suggestions.js
• scripts/find-skill-upgrade-candidates.js
• scripts/build-memory-patch-candidates.js
• scripts/build-audit-report.js
• scripts/build-dashboard.js

Setup

When this skill is used to initialize a workspace, immediately ensure these three recurring jobs exist (create missing ones, update same-name jobs if they already exist):

• - shared-memory-os daily maintenance — daily light maintenance
• shared-memory-os weekly review — weekly full maintenance + dashboard/audit summary
• shared-memory-os monthly deep maintenance — monthly deep cleanup and review

Recommended default schedules (Asia/Shanghai):

• - daily: 15 3
• weekly: 30 3 1
• monthly: 45 3 1 *

Recommended job shape:

• - sessionTarget: isolated
• payload.kind: agentTurn
• payload.toolsAllow: [exec, read]
• delivery.mode: none
• failure alert enabled after 1 failure

After cron jobs are ensured, run one full maintenance pass so reports exist immediately.

For first-time setup from a terminal instead of an agent session, run node skills/shared-memory-os/scripts/bootstrap-shared-memory-os.js or bash skills/shared-memory-os/scripts/bootstrap-shared-memory-os.sh from the workspace root. These are fallback entrypoints for manual/bootstrap use.

If you only need to repair or reapply automation from a terminal, run node skills/shared-memory-os/scripts/ensure-shared-memory-crons.js or bash skills/shared-memory-os/scripts/ensure-shared-memory-crons.sh. The installer is intended to be idempotent: same-name jobs are updated, missing jobs are created.

Guidance

中文

Shared Memory OS 的目标，是把 workspace 里的共享记忆变成“可治理、可维护、可进化、可协同”的系统，而不是零散笔记。

核心能力

• - 分层记忆治理
• 主动沉淀 .learnings/
• review cadence 与 archive strategy
• 冲突处理顺序
• 健康评分与历史趋势
• 带 evidence/confidence 的重复经验检测
• 带 review 状态的 promotion candidate 检测
• promotion 建议与 MEMORY.md patch 候选生成
• 周报、审计报告、dashboard 生成
• 带生命周期提示的 stale durable memory 检测
• 反复验证规则检测
• 低价值 learning 检测
• 工作流优化建议生成
• 冲突日志与冲突审查报告
• learning 合并建议
• skill 升级候选检测
• 初始化与迁移脚本
• 自学习进化闭环
• 一键 maintenance runner
• policy profile / workspace profile
• 与其他 skills 的协同治理
• 多 agent 写入与审核边界建议
• 周期性 maintenance day 指南

参考文档

• - references/zh-cn-guide.md
• references/governance-guide.md
• references/pattern-harvesting.md
• references/archive-strategy.md
• references/learnings-template.md
• references/health-check.md
• references/self-improving-loop.md
• references/health-score-model.md
• references/conflict-template.md
• references/migration-guide.md
• references/policy-profiles.md
• references/workspace-profiles.md
• references/cross-skill-collaboration.md
• references/multi-agent-governance.md
• references/maintenance-day.md

脚本

• - scripts/bootstrap-shared-memory-os.js
• scripts/ensure-shared-memory-crons.js
• scripts/init-shared-memory-os.js
• scripts/migrate-notes-into-memory.js
• scripts/run-shared-memory-maintenance.js
• scripts/check-memory-health.js
• scripts/rebuild-learnings-index.js
• scripts/find-duplicate-learnings.js
• scripts/find-promotion-candidates.js
• scripts/find-stale-durable-memory.js
• scripts/find-validated-rules.js
• scripts/find-low-value-learnings.js
• scripts/build-workflow-optimization-suggestions.js
• scripts/log-memory-conflict.js
• scripts/build-conflict-review-report.js
• scripts/build-weekly-review.js
• scripts/record-health-snapshot.js
• scripts/build-promotion-suggestions.js
• scripts/build-learning-merge-suggestions.js
• scripts/find-skill-upgrade-candidates.js
• scripts/build-memory-patch-candidates.js
• scripts/build-audit-report.js
• scripts/build-dashboard.js

初始化方式

当此 skill 用于初始化 workspace 时，应立即确保以下 3 个同名任务存在（已存在则更新，不存在则创建）：

• - shared-memory-os daily maintenance
• shared-memory-os weekly review
• shared-memory-os monthly deep maintenance

推荐默认时区与时间（Asia/Shanghai）：

• - daily: 15 3
• weekly: 30 3 1
• monthly: 45 3 1 *

推荐任务形态：

• - sessionTarget: isolated
• payload.kind: agentTurn
• payload.toolsAllow: [exec, read]
• delivery.mode: none
• 开启 failure alert，连续 1 次失败就报警

确保 cron 任务后，应立即跑一次完整 maintenance，让 reports 立刻可用。

如果是在终端里手动初始化，而不是在 agent 会话里，使用：node skills/shared-memory-os/scripts/bootstrap-shared-memory-os.js 或 bash skills/shared