agent-infra

Record and Replay AI Agent Trajectories.

MCP server for structured AI reasoning — decompose problems into atomic thoughts with dependency graphs, confidence tracking, and interactive D3.js visualization

Real-time Claude Code session analytics — time breakdown, cost tracking, stuck loop detection, and live dashboard

An Agent Infra Benchmark Suite for Agentic Tool Serving (AgentCore, Lambda, GCP, AgentRun, E2b etc.)

CLI proxy that reduces LLM token consumption by 60-90% on common dev commands. Single Rust binary, zero dependencies

Lightweight headless browser for AI agents

Lightweight, auditable intent router for LiteLLM and OpenAI-compatible model gateways.

The agent-run work-unit CLI: capability-scoped agents whose every privileged effect is verified against its signed owner, driving a work unit through one signed pipeline to a merged PR.

self-hosted sandbox runtime for Agents with elasticity and strong filesystem state

Agent-native task control plane for humans, agents, tools, artifacts, and approvals.

Front Desk projection + sync room for bounded-systems (org project #2)

Guest-agnostic room+door capability runtime — the core library claude-box is built on.

AgentTeams 动作控制平面：精确授权、崩溃对账与包外验真

Three agents, one shared note store, and a human who has to say yes. Minimal multi-agent memory loop with the cogram.keyword_recall skill (GOAI 2026 Agent Infra, educational slice).

A capability-secured box for agent sessions — its authority is the door references it holds (keeper/scout/concierge/net), parent-agnostic.

Switchback Governance · 折返治理 — the human-in-the-loop governance layer for multi-agent teams (GOAI 2026 Track 1)