Privacy Concierge

Personal AI assistant that monitors your online privacy, calculates exposure scores, automates data broker opt-outs, tracks breaches, and offers privacy advice.

Do not install or provide secrets yet. Ask the publisher for clarification and fixes first: 1) Explain why the registry metadata lists no env vars while SKILL.md requires GROQ_API_KEY, SUPABASE_URL/SUPABASE_ANON_KEY, and TELEGRAM_BOT_TOKEN; update registry metadata to match. 2) Remove or restrict tools/read.js (or replace it with a limited-purpose API) — a generic file-read helper can be used to access local secrets and is unnecessary for the described features. 3) Clarify which Supabase key is required; for deletions/DSARs a service-role key is more likely needed (and is sensitive); prefer least privilege and explicit guidance on key scope. 4) Explain how email sending is implemented and which credentials are required (SMTP/API keys) and update SKILL.md. 5) Ask for the upstream source or homepage and a code audit; run the skill in a sandbox with monitored network access before giving it real credentials. If you must try it, create dedicated, limited-permission test credentials (test Supabase project, Telegram bot scoped to a test chat, and a throwaway LLM key) and do not reuse any production secrets.

Type: OpenClaw Skill Name: privacy-concierge Version: 1.0.0 The skill bundle contains a critical vulnerability in `tools/read.js`. This file exposes a `readFile` function that directly uses `params.path` in `fs.readFileSync`, allowing the AI agent to read arbitrary files on the system. This lack of input sanitization or access control makes the agent highly susceptible to prompt injection attacks, potentially leading to sensitive information disclosure (e.g., credentials, private keys) or further system compromise, despite the stated benign purpose in `SKILL.md` and other documentation.