Look

What to consider before installing: - Cron and self-update: The skill asks the agent to register a daily cron job that pulls code from GitHub. Automatic self-updates can change behavior after installation; only accept this if you trust the upstream repo and are comfortable with scheduled remote fetches. Prefer requiring manual updates or code signing verification. - Cross-skill writes: The instructions create/write files under ~/openclaw/db/ocas-elephas/intake/, which is outside the declared filesystem permissions. Confirm you want this skill to write into another skill's intake directory (that will propagate extracted entities to other components). - Sensitive data: The skill will ingest images and EXIF (location/time). Even though it doesn't request credentials, it will store journals and artifacts locally; review retention settings in config.json (default 30 days). If you handle sensitive images, audit where journals/artifacts are written and consider stricter retention. - External code fetch: Because the package is instruction-only but references a GitHub repo and self-update, the agent may fetch code at runtime that wasn't reviewed. If you want to proceed, inspect the referenced GitHub repo (https://github.com/indigokarasu/look) to verify source and review update mechanism. - Manifest mismatch: Ask the publisher to update skill.json to declare any additional filesystem paths (elephas intake) and to explicitly describe and optionally disable the auto-update cron. That alignment would reduce ambiguity and make the skill closer to benign. What would change this assessment to 'benign': a manifest that explicitly declares all filesystem writes (including elephas intake), a documented, auditable, and optional update mechanism (no silent cron installs), and either inclusion of the code for static review or a signed, verifiable update process.