AI 执行与监察体系（Agent Oversight Framework）：分层治理规范 + L2 运行时监察参照实现——证据分级、自校正、边界诚实。A layered governance spec for AI agents, plus a runnable reference implementation of the L2 runtime-oversight layer. Docs CC BY-SA 4.0 / code Apache-2.0.
VeriClaw 爪印: Apple-native AI agent correction and OpenClaw companion workflow for LLM supervision, hallucination remediation, and role-drift diagnosis.
The control plane for AI desktop agents: watch, gate, and audit any agent before it touches a real machine. Sub-ms policy engine, independent AI verifier, human-in-the-loop gate, tamper-evident audit log, and a live Control Hub. Local-first and model-agnostic.
Human-in-the-loop console for autonomous pentest/red-team agents — review live agent activity and approve or deny high-risk actions before they run.
Measuring computer-use agent oversight when the overseer cannot see the screen
Active oversight layer (cockpit) for Claude Code agents — NOT an agent itself. Forces plan-first, surfaces decisions, makes thinking visible. Middle ground between yolo and baby-every-command.
Measure how, not just whether
To associate your repository with the agent-oversight topic, visit your repo's landing page and select "manage topics."