Unified Memory V5

Provides a layered memory management system with atomic transactions, smart deduplication, vector search optimization, and lifecycle management for consisten...

This bundle appears to implement a full-featured memory system, but there are several red flags and inconsistencies you should address before installing: - Don’t trust metadata alone: the package claims no env/config requirements but the README/SKILL.md and source reference many env vars and config files (e.g., OLLAMA_URL, UNIFIED_MEMORY_*, ~/.unified-memory, ~/.openclaw). Expect to be prompted to provide these and the code will read/write to your home directory. - Prompt-injection risk: the SKILL.md contains patterns that look like attempts to override prompts, and package.json declares a before_prompt_build hook. If you enable this skill it could modify agent prompts or behavior — only allow it if you fully trust the author. - Code provenance and install safety: there is no explicit install spec but package.json lists many dependencies. Inspect the repo, lockfile, and large merged-src.js for any network calls, hidden endpoints or exec() usage. Prefer running in an isolated environment (container or VM) first. - Operational caution: do not run installation or deploy scripts (deploy-atomic-fixes.sh, verify-repairs.sh, or any scripts under ./scripts) on a production system without auditing them. These scripts may perform filesystem operations. Actionable steps before installing: 1) Review the source (merged-src.js, scripts) for any outbound network calls, eval/exec usages, or unexpected endpoints. 2) Run the package inside a sandboxed container with minimal privileges and no access to secrets. 3) If you need its functionality, prefer installing from a verified upstream repo with reproducible builds and pinned dependencies. 4) If you enable hooks, audit what data they receive and ensure they cannot alter system-level prompts. If you cannot perform these checks or do not trust the author (github: mouxangithub appears in docs but source registry owner is unknown), treat this skill as risky and avoid installing it in production or on machines with sensitive data.

Type: OpenClaw Skill Name: unified-memory-v5 Version: 5.2.4 The Unified Memory skill bundle is a highly sophisticated and well-documented memory management system for AI agents. It implements advanced features such as hybrid search (BM25 + Vector + RRF), atomic transactions with Write-Ahead Logging (WAL) for data consistency, and a multi-layered memory architecture (L0-L3). The documentation (SKILL.md, README.md, and merged-docs.md) is technically consistent and describes legitimate knowledge management workflows. No evidence of malicious intent, data exfiltration, unauthorized remote execution, or harmful prompt injections was found. The requested filesystem permissions are appropriately restricted to the agent's workspace directory (~/.openclaw/workspace/memory/).