I built a small open-source CLI for a problem I kept seeing in AI-built Supabase apps: the agent working in the repo has far more power than anyone intended.
It reads your repo (and optionally ~/.claude.json and ~/.cursor/mcp.json) and reports:
service_role / sb_secret_ keys an agent can read, or that ship to the browsersbp_ access tokens and DB URLs with passwordsread_only=true or project_refsupabase db reset --linked or supabase projects deleteIt is read-only: no network, no database, and it never prints a secret. It does not check RLS. For that I point people to supabase db advisors --type security and rlsgrid.
Repo: https://github.com/karthikkushi/karchi-guard
I would like feedback on the checks. Are there agent setups I am missing?
(I built this with help from Claude.)
Karthikein M R introduces Karchi Guard, a CLI tool designed to identify potential security issues in AI-built Supabase applications. The tool checks for exposed service roles, access tokens, and other vulnerabilities without altering data. Karthikein seeks feedback on the tool's checks and any potential agent setups that might be missing.