memory-orchestrator

Layered memory orchestration for OpenClaw conversations. Use when implementing or maintaining a memory system that must classify user input by domain, captur...

This skill appears to do what it claims: manage a local, white-box memory directory and provides scripts to gate, capture, recall, and reflect. Before installing or enabling it widely: 1) Confirm where memory will be stored — set MEMORY_ROOT explicitly to a safe, isolated workspace directory (the default is ./memory in the agent's working directory). 2) Review and monitor the created memory/ files (topics, objects, daily, reflections, indexes) because the skill will write user-provided text to disk and can materialize 'preferences' and 'decisions' files automatically. 3) If you allow autonomous invocation, be aware the skill may write files whenever its gate triggers; if that is undesirable, require user-invocation only. 4) Avoid setting MEMORY_ROOT to any sensitive system path and prefer running the skill in an isolated container/workspace. 5) If you need networked or multi-tenant deployment, audit the code further (it currently performs no network calls but writes files locally).

Type: OpenClaw Skill Name: mem-orchestrator Version: 1.0.0 The Memory Orchestrator bundle is a well-structured system for managing persistent agent memory using local YAML and Markdown files. The Python scripts (such as apply_memory_events.py and memory_cli.py) handle data storage, retrieval, and indexing within a user-defined directory without any network activity or unauthorized file access. Security-wise, the system includes proper filename sanitization (slugify) to prevent path traversal and uses standard subprocess orchestration for its internal components. The SKILL.md instructions are entirely aligned with the stated purpose of memory management and do not contain any malicious prompt injection attempts or instructions to exfiltrate data.