Shepherd: A Runtime Substrate Empowering Meta-Agents with a Formalized Execution Trace
Provides an efficient infrastructure for programming meta-agents, enabling runtime intervention, counterfactual optimization, and tree-RL training.
Shepherd introduces a functional programming model that formalizes meta-agent operations with a Git-like execution trace, enabling fast forking and replay. It achieves 5× faster forking than Docker, >95% prompt-cache reuse on replay, and improves pair coding pass rates from 28.8% to 54.7% on CooperBench.
We introduce Shepherd, a functional programming model that formalizes meta-agent operations on target agents as functions, with core operations mechanized in Lean. Shepherd records every agent-environment interaction as a typed event in a Git-like execution trace, enabling any past state to be forked and replayed. The system forks the agent process and its filesystem $5\times$ faster than Docker, achieving $>95\%$ prompt-cache reuse on replay. We demonstrate the model through three applications. First, in runtime intervention, a live supervisor increases pair coding pass rates from 28.8% to 54.7% on CooperBench. Second, in counterfactual meta-optimization, branching exploration outperforms baselines across four benchmarks by up to 11 points while reducing wall-clock time by up to 58%. Third, in Tree-RL training, forking rollouts at selected turns improves TerminalBench-2 performance from 34.2% to 39.4%. These results establish Shepherd as an efficient infrastructure for programming meta-agents. We open-source the system to support future research.