向量记忆自我进化系统

向量记忆自我进化系统 - 结合 BGE 向量模型、Chroma 向量库、四层记忆架构，实现自动错误捕获、用户纠正学习、最佳实践积累、语义检索的自我进化能力。

This skill appears to implement a local vector-memory system (BGE embedding service + Chroma DB) and mostly restricts activity to your home workspace and localhost. Before installing or enabling auto-capture: 1) Verify the missing files and functions: SKILL.md references setup_memory_system.sh, start_bge_service.sh, and scripts/code_security_scan.py that are not in the package, and redact_tool calls memory_api.log which does not exist — these will cause runtime errors and may leave redaction/scanning nonfunctional. 2) Confirm the origin of the BGE embedding service you will run on localhost (who provides it, and whether it sends data externally). 3) Inspect and test the code in a sandbox or VM (especially vectorize_memories.py which will POST memory text to the embedding service) before enabling cron/auto-capture. 4) Because the skill will read and write files under ~/.openclaw/workspace and could store extracted content, avoid feeding it sensitive secrets until you confirm redaction actually works. 5) If you want to proceed, request the missing scripts or a corrected release (fix the memory_api.log reference and supply the setup/start/scan scripts) or run the bundled scripts manually after review.

Type: OpenClaw Skill Name: vector-memory-self-evolution Version: 2.1.1 The bundle implements a sophisticated four-layer memory system (L1-L4) for OpenClaw agents, utilizing ChromaDB and BGE embeddings for semantic search and 'self-evolution' through error and correction tracking. It includes a dedicated redaction tool (`redact_tool.py`) to strip sensitive information like API keys, tokens, and passwords from logs before storage, demonstrating a security-conscious design. While the system possesses high-risk capabilities such as modifying its own rule files (`SOUL.md`) and capturing command history, these behaviors are transparently documented and aligned with the stated goal of improving agent performance over time. The code lacks indicators of malicious intent, data exfiltration, or unauthorized remote access, and it explicitly mentions the removal of previous network redirections (HF_ENDPOINT) in its changelog.