Bug

gbrain config set openai_api_key sk-... confirms by printing the full key in plaintext to stderr:

Set openai_api_key = sk-svcacct-MGoTY7Gw3zMRXJb7n_7Y...full-key-visible...

The same applies to any sensitive key set via gbrain config set.

Why this matters

• Keys end up in shell history (if user ran the command directly)
• Keys end up in CI logs / agent transcripts (if a tool wrapper called it)
• Keys end up in tmux/screen scrollback
• Keys end up in any log capture that sucks in stderr

In my session, an AI agent helping the user set up gbrain leaked their full OpenAI service-account key to the conversation transcript via this confirmation message. The user now has to rotate the key.

Suggested fix

Redact the value in confirmation output. Mask all but first 8 + last 4 chars, or just print Set openai_api_key = <REDACTED>. Same treatment for any key whose key-name matches *_api_key, *_token, *_secret, *password*, *pat*.

For comparison: claude mcp add and the gstack gstack-gbrain-supabase-provision helper both correctly redact their secret args.

Repro

gbrain config set openai_api_key sk-test-1234567890abcdef
# Outputs: Set openai_api_key = sk-test-1234567890abcdef

Environment

gbrain 0.32.0