shared-memory-governor

Govern a file-based shared-memory layer for OpenClaw multi-agent and subagent systems. Preserve each agent’s private memory while adding a separate, reviewab...

This is an instruction-only governance skill (no code, no installs, no credentials requested) and appears coherent with its purpose. Before enabling it in your workspace: 1) confirm the intended sharedRoot path — the reference defaults to /root/.openclaw/shared-memory, so change it if your workspace is elsewhere; 2) review schedule settings in the example config (privateMaintenance is enabled in the example) and disable any recurring tasks you do not want running; 3) ensure local SHARED_ATTACH.md and AGENTS.md guidance are accurate for each agent so the read/attach rules are enforced; and 4) remember this is documentation for operator behavior — the safety guarantees rely on the agent implementation following these rules, so audit any runtime that will implement these steps if you need stronger enforcement (e.g., code-level guards or OS permissions).

Type: OpenClaw Skill Name: shared-memory-governor Version: 1.0.2 The 'shared-memory-governor' skill is a governance framework designed to manage shared data layers across multiple AI agents within the OpenClaw environment. The skill's logic, defined across SKILL.md and several reference documents, focuses on maintaining strict identity isolation and preventing the accidental promotion of private data or secrets into shared space. It includes explicit safety boundaries that forbid reading credentials, SSH keys, or browser data. While it performs file modifications (e.g., updating AGENTS.md for startup guidance) and manages task schedules for maintenance, these actions are transparently documented and strictly scoped to the stated purpose of memory management, with no evidence of malicious intent, data exfiltration, or harmful prompt injection.