Self-improving Agent Memory Upgrade (SurrealDB)

A comprehensive knowledge graph memory system with semantic search, episodic memory, working memory, automatic context injection, and per-agent isolation.

This skill implements a powerful local memory system but includes several high-risk operations that you must consciously opt into. Before installing or enabling features: - Audit your memory files (MEMORY.md and memory/*.md) and remove any secrets (API keys, passwords, private tokens). The extraction job sends those files to OpenAI. - Prefer manual installation of SurrealDB; avoid the curl | sh installer unless you fully trust the source. The installer is documented as opt-in but the gateway exposes an 'install' action that will run it if you click it. - Keep SurrealDB bound to 127.0.0.1 and change default credentials (root/root) before exposing any network access. - Do not enable auto-injection until you review the knowledge graph contents and start with a small max_facts limit (3–5). Auto-injection modifies the system prompt and can significantly change agent behavior. - Verify how cron jobs run on your platform: the changelog claims jobs run in isolated background sessions, but some documentation (SECURITY.md) still describes main-session cron behavior. Confirm the current runtime behavior (sessionTarget) before enabling scheduled extraction or relation discovery. - If you plan to apply the code patches to OpenClaw, run the provided integrate-openclaw.sh in dry-run mode and keep a git backup of your OpenClaw source; only apply with --apply after manual review. Given these caveats, the skill is coherent with its purpose but carries non-trivial operational and privacy risks; proceed only after the checks above.

Type: OpenClaw Skill Name: surrealdb-knowledge-graph-memory Version: 2.3.0 This skill bundle implements a complex knowledge graph memory system with several high-risk capabilities that, while documented and aligned with its stated purpose, present a significant attack surface. Key indicators include 'scripts/integrate-openclaw.sh' which uses sed to patch the host application's source code (opt-in via --apply), 'scripts/install.sh' which contains an optional curl|sh network installer, and 'scripts/extract-knowledge.py' which reads workspace files and sends their content to the OpenAI API. Additionally, the skill registers multiple cron jobs for background extraction and relationship discovery, and performs automatic context injection into agent prompts, although it includes a security filter in 'scripts/mcp-server-v2.py' to prevent the injection of sensitive strings like API keys or passwords.