uber/ADR

git clone https://github.com/uber/ADR && cd ADR/Detection && uv sync
export ANTHROPIC_API_KEY="..." OPENAI_API_KEY="..."

Usage

• •Run the default detector: `adr` (ADR dual-agent); keyless smoke tests use --detector llamafirewall
• •Follow docs/REPRODUCIBILITY.md to inflate the packed benchmark, run detectors, and plot figures
• •Deploy the ADR Sensor to collect and normalize agent telemetry from Claude Code, Cursor, Codex, and others

Key features

• ✓ADR Observability: captures agent intent, tool use, and execution traces across 7+ AI coding tools on macOS, Linux, and Windows plus internal/customer-facing agents
• ✓ADR-Bench: 300+ tasks, 133 MCP servers, coverage of all 17 agent attack techniques
• ✓ADR Detection: two-tier architecture combining high-recall triage with deeper agentic reasoning on suspicious sessions
• ✓ADR Prevention: stops unsafe actions before they cause harm (not included in the current open-source release)

Best for

Security teams securing enterprise employee-facing AI coding agents (Cursor, Claude Code, Codex) and customer-facing support agents.

Caveats

• ⚠ADR Prevention component is not in the open-source release (stay tuned)
• ⚠The offline ADR Explorer red-teaming engine is not included in this repo
• ⚠Detection/ contains synthetic benchmark fixtures (fake credentials, emulated environments, prompt-injection scenarios) for defensive security research only