SAGE: an RL-driven self-improving skill library¶

Fidelity: core mechanism. The deterministic mini-suite does not stand in for the paper's full benchmark.

Paper information¶

Field	Details
Paper	SAGE: an RL-driven self-improving skill library (arXiv 2512.17102)
Organization	University of Wisconsin–Madison / Amazon Web Services
First public release	2025-12-18 (arXiv v1)
Authors' code	No official author repository was found.
Adapter / method	`sage`
Local reproduction code	`src/auto_research/agent_research/`

Original paper¶

Motivation and method¶

Abstract skills from successful trajectories, revise or retire them after failures, and use task rewards to learn retrieval and reuse.

flowchart LR
 A["Public input or history"] --> B["SAGE training and execution loop"]
 B --> C["Paper-specific state or objective"]
 C --> D["Matched-budget evaluation"]

Key figure from the paper¶

Figure 5: Training flow and key optimization steps. From the original paper; copyright remains with the authors.

Core equations¶

\[ s^*=\arg\max_{s\in\mathcal S}q_\phi(s|x),\quad\mathcal S\leftarrow\operatorname{Update}(\mathcal S,\tau,R),\quad\max_\phi\mathbb E[R]. \]

Paper results¶

On continual tasks, the paper reports improved success and sample efficiency from skill reuse. No production online A/B test is reported; no online figures are inferred.

Local reproduction¶

PlanBench mini-suite, 120 episodes, seed 42: joint success 1.0000, average cost 0.3000, and nonzero method-specific telemetry.

auto-research agent-eval --method sage --benchmark planbench-mini --episodes 120 --seed 42
auto-research evolve --model agent --dataset planbench-mini --direction "组合 sage 与已安装论文算子" --generations 2 --population 4

See the fixed metrics.

Reproduction boundaries¶

The local run checks the paper-specific objective, state updates, and fair budgets. It does not reproduce large models, multi-GPU RL, private environments, real websites, or the full benchmark. It verifies mechanisms rather than claiming numerical reproduction of the original tables.