Taste

Generates personalized recommendations from real consumption data by scanning email/calendar, enriching venues, and explaining suggestions with prior behavior.

Before installing or enabling this skill, get answers and make changes to reduce risk: 1) Ask the author/platform how email/calendar access is granted and scoped (what OAuth scopes, consent screens, and tokens are used?) and insist credentials are explicit and limited. 2) Confirm how Google Maps enrichment is performed (official API with an API key vs. scraping public pages) and where any API keys would be stored. 3) Ask whether the skill will automatically register a cron job or auto-update itself; if so, require explicit user approval for updates and a way to disable cron/self-updates. 4) Understand where extracted data is stored, retention policy, encryption at rest, and how to delete data. 5) If you cannot verify these behaviors, consider running in a restricted sandbox or deny email/calendar access and use manual signal ingestion instead. 6) Prefer skills that declare required credentials and install steps explicitly; lack of declared credentials for email/calendar/Google Maps is a red flag. If the platform provides built-in connectors for email/calendar/maps, confirm that the skill will use those connectors without requiring additional long-lived secrets.

Type: OpenClaw Skill Name: ocas-taste Version: 3.0.1 The skill implements a high-risk self-update mechanism in SKILL.md (the `taste.update` command) that downloads and overwrites its own files from a remote GitHub repository using shell commands, creating a significant path for remote code execution (RCE) if the source is compromised. Additionally, the skill is designed to scan the user's private email and calendar for sensitive transactional data from various services (Amazon, DoorDash, hotel bookings), as detailed in references/email_extraction.md. While these behaviors are documented and aligned with the stated goal of personalized recommendations, the combination of broad sensitive data access and an unverified remote update path presents a substantial security risk.