High: setup.py exposes API keys during interactive configuration

[tg12]

Severity: High

Evidence:

• scripts/setup.py:46-51 uses input() for prompts, so values typed for secret prompts are echoed by the terminal.
• scripts/setup.py:178-186 asks for OPENAI_API_KEY, ANTHROPIC_API_KEY, and GEMINI_API_KEY values through that echoed prompt path.
• scripts/setup.py:60-89 writes KEY=value to .env, then prints the full Set {key}={value} string on stdout.
• .env.example:33-39 documents those values as API keys.

Reproduction:

1. Run python3 scripts/setup.py.
2. Choose provider openai, anthropic, or gemini.
3. Enter any token-like value, for example sk-test-redacted.
4. Observe that the value is visible while typing and then printed again as Set OPENAI_API_KEY=sk-test-redacted.

Impact:
API keys are exposed in terminal scrollback, shell/session recording, CI logs if setup is scripted, and AI-agent transcripts. This is especially risky for a repo designed to be operated by coding agents.

Smallest credible fix:

• Use getpass.getpass() for secret prompts.
• Treat keys matching *_API_KEY, *_TOKEN, *_PASSWORD, and similar names as sensitive in _patch_env() and print only a masked confirmation such as Set OPENAI_API_KEY=<redacted>.
• Avoid echoing existing secret values during reconfiguration.
• Add a small test around _patch_env() output redaction for sensitive keys.