Huo15 Knowledge Base

This skill appears to do what it claims: create per-agent knowledge bases and use an LLM to compile and lint entries. Important cautions before installing/using: - Data exposure: compile and lint build prompt files that include the full contents of your raw documents/wiki entries and instruct the environment to send them to the configured LLM provider. Do NOT ingest sensitive or confidential documents unless you trust the configured provider and understand its data handling. - Credentials use: kb-llm.py reads models.json under your agent directories to discover provider baseUrl and apiKey and will use those credentials to call the provider. Confirm models.json points to a trusted provider and does not contain unexpected endpoints. - Bulk activation: scripts/install-all-agents.sh will create kb/ and config.json across all agents under ~/.openclaw/agents — review that script before running as it will write into every agent workspace. - Test in isolation: activate the skill for a single test agent first (or run scripts in a sandbox), verify the provider and outputs, and confirm no sensitive data is being sent. - Portability note: some scripts use sed -i '' (BSD syntax) which may behave differently on Linux; this is not a security issue but could affect operation. If you want, I can: (1) point out the exact lines where documents are concatenated into prompts, (2) show how to edit the scripts to avoid sending raw content (e.g., only send metadata or excerpts), or (3) produce a safe checklist to run before using the skill in production.