Abstract: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.
| Comments: | 56 pages, 21 figures, 14 tables |
| Subjects: | Artificial Intelligence (cs.AI); Programming Languages (cs.PL); Software Engineering (cs.SE) |
| Cite as: | arXiv:2605.10913 [cs.AI] |
| (or arXiv:2605.10913v1 [cs.AI] for this version) | |
| https://doi.org/10.48550/arXiv.2605.10913 arXiv-issued DOI via DataCite (pending registration) |
Submission history
From: Simon Yu [view email]
[v1]
Mon, 11 May 2026 17:50:51 UTC (8,533 KB)