persona-knowledge

Persistent, incremental, searchable persona knowledge base. Ingests data from Obsidian vaults, chat exports, X/Twitter archives, and more into a MemPalace-ba...

This skill appears to do what it claims: ingest local archives and build a local persona knowledge base. Before installing/using it, consider: 1) You will be giving the skill access to local files (chat exports, Obsidian vaults, social archive directories) and it will store backups and indexes under ~/.openpersona/knowledge — do not ingest sensitive data you don't want preserved in plain files. 2) The iMessage adapter reads the Messages SQLite DB (macOS Full Disk Access may be needed) — grant permissions only if you trust the code. 3) It depends on the third‑party mempalace library (pip install mempalace); review that library if you need assurance about where vector indices are stored or whether it contacts external services. 4) The agent is allowed to write wiki pages and (per SKILL.md) may use Bash and WebSearch if the platform exposes those tools — only enable autonomous invocation if you trust the agent's behavior and limit sources you provide. 5) Use --dry-run first to verify parsed messages and PII flags, and inspect ~/.openpersona/knowledge before running exports. If you want stronger protections, run the ingestion in an isolated environment, avoid ingesting highly sensitive files, and/or encrypt the knowledge directory.

Type: OpenClaw Skill Name: persona-knowledge Version: 0.2.2 The persona-knowledge skill bundle is designed to aggregate highly sensitive personal data, including iMessage databases (chat.db), social media archives, and various chat exports, to create a persona training dataset. Key indicators of risk include the chat_export.py adapter, which provides logic for reading private SQLite databases, and the ingest.py script, which contains a PII scanner specifically targeting SSNs, credit card numbers, and passwords. While these capabilities are plausibly required for the stated purpose of building a comprehensive knowledge base, the automated processing of such high-value sensitive data and the requirement for broad tool permissions (Bash, Read, Write, WebSearch) represent a significant security surface. No evidence of intentional data exfiltration or malicious persistence was found.