Add two_brains - AI agent with dual safety critic

[kholovmasrur2007-netizen]

What this adds

Adds two_brains to the Open-source projects section, alphabetically between Teenage AGI and UFO.

What it is

two_brains is an open-source 3-brain AI agent (planner → critic → executor). The interesting bit is the dual critic: a deterministic regex-driven SafetyCritic runs alongside the LLM critic, and the combined score is min(both). Even if the model is 100% confident the plan is fine, the safety critic can drag it under the 85-point execution bar — so rm -rf, sudo, chmod 777, curl | wget, sandbox-traversal, fuzzy success criteria, and writes-without-declared-risks all get blocked before the executor ever touches the disk.

Other safety primitives:

• Per-user sandbox with parametrised traversal tests (/etc/passwd, C:/Windows, ..\escape, multi-level escapes)
• 300 s wall-clock timeout in the agent loop
• Disk-space pre-flight on write_file
• WebSocket per-IP rate limit (slowapi only covers REST)
• JWT auth, daily per-user quotas, audit log, Prometheus metrics
• Web UI streams every pipeline phase live (8 events: planner_started → plan_ready → critic_started → critique_ready → executor_started → tool_call → tool_result → done)
• Telegram bot front-end included
• Providers: deterministic, mock, Anthropic, OpenAI, agent (Claude tool-use), local-agent (offline pattern matcher)

Stats

• 238 tests, CI green on Python 3.11 + 3.12
• MIT-licensed
• One-line install: curl -fsSL https://raw.githubusercontent.com/kholovmasrur2007-netizen/two_brains/main/install.sh | bash
• Latest release: v3.0.0

Happy to amend the entry if you'd like a different category, shorter description, or a different placement in the list.