BEE Belief Extraction Engine
PassAudited by ClawScan on May 10, 2026.
Overview
This skill is coherent for installing a persistent OpenClaw memory extension, but users should understand it installs external code and stores LLM-derived beliefs across sessions.
This appears to be a purpose-aligned persistent memory setup skill. Before installing, verify the external package source, decide whether indefinite belief storage is acceptable, and understand that session-derived information may be processed by the configured LLM provider.
Findings (3)
Artifact-based informational review of SKILL.md, metadata, install specs, static scan signals, and capability signals. ClawScan does not execute the skill or run runtime probes.
Installing the package gives external code the ability to run as part of the OpenClaw extension environment.
The skill directs users to install external package code globally, while the provided artifact set contains only SKILL.md and no package code for review.
npm install -g @skysphere-labs/openclaw-bee ... npm install -g github:skysphere-labs/openclaw-bee
Install only if you trust the npm/GitHub source; consider pinning a version and reviewing the package repository before enabling it.
Sensitive or incorrect beliefs could persist and be reused later, affecting future agent behavior.
The skill intentionally creates persistent memory that can influence future sessions and retain session-derived information indefinitely.
Extracts beliefs at session end ... Injects recalled context before every agent spawn ... Beliefs live in a SQLite database (`vector.db`) and persist indefinitely across restarts.
Use separate agentId namespaces, review or delete the SQLite database when needed, and disable extraction if you do not want persistent memory.
Information from sessions may be sent to the configured model provider for belief extraction.
Belief extraction uses an external model/provider path, which may process session-derived content. This is disclosed and purpose-aligned, but the artifact does not detail provider-side handling.
Extracts beliefs at session end via a lightweight LLM call (Haiku by default) ... `extractionModel`: `anthropic/claude-haiku-4-5`
Confirm your OpenClaw model-provider privacy settings and disable or change `extractionModel` if provider processing is not acceptable.