Unsupervised Skill Discovery for Agentic Data Analysis

2026-05-03 · Source: cs.AI updates on arXiv.org · Field: Technology & Digital — Artificial Intelligence & Machine Learning, Data Science & Analytics · Depth: Expert, extended

Summary

DataCOPE is an unsupervised verifier-guided skill discovery framework designed to enhance data-analytic agents by injecting reusable procedural knowledge without updating model parameters. It addresses the challenge of discovering effective data analysis skills from unlabeled exploration by iteratively coordinating a Data-Analytic Agent for trajectory generation, an Unsupervised Verifier for signal extraction, and a Skill Manager for contrastive skill distillation. For report-style analysis, DataCOPE employs an Adaptive Checklist Verifier that refines task-specific criteria and scores reports by verifiable coverage. For reasoning-style analysis, it uses an Answer Agreement Verifier that groups trajectories by answer agreement and leverages self-consistency. Evaluated on Deep Data Research (report-style) and DABStep (reasoning-style), DataCOPE consistently improved held-out performance. Across four matched base models (Claude-Sonnet-4.6, Claude-Sonnet-4.5, GPT-5.2, DeepSeek-V4-Pro, Qwen3.5-397B), it boosted mean scores by 9.71% on report-style tasks and 32.30% on reasoning-style tasks, outperforming baselines like Anthropic's Skill Creator.

Key takeaway

For Machine Learning Engineers developing data-analytic agents, DataCOPE offers a robust method to enhance agent performance without costly manual supervision. You should consider implementing its unsupervised verifier-guided skill discovery framework to distill reusable procedural knowledge from unlabeled exploration. This approach significantly improves accuracy on both report-style and reasoning-style tasks, reducing token consumption and enabling more efficient, generalizable agent behavior across diverse models.

Key insights

DataCOPE enables unsupervised skill discovery for data-analytic agents using verifier-guided iterative refinement of exploration trajectories.

Principles

• Unsupervised verifier signals are crucial for effective skill discovery.
• Iterative refinement improves skill generalization and robustness.
• Appropriate skill granularity is essential for balancing specialization and generalization.

Method

DataCOPE iteratively coordinates a Data-Analytic Agent, an Unsupervised Verifier (Adaptive Checklist or Answer Agreement), and a Skill Manager to distill reusable procedures from grouped exploration trajectories.

In practice

• Implement an Adaptive Checklist Verifier for open-ended report generation tasks.
• Use an Answer Agreement Verifier with self-consistency for fixed-answer reasoning tasks.
• Filter trajectories with saturated self-consistency to focus skill refinement on uncertain cases.

Topics

• Unsupervised Skill Discovery
• Data-Analytic Agents
• Large Language Models
• Adaptive Checklist Verifier
• Answer Agreement Verifier
• Deep Data Research
• DABStep

Code references

• anthropics/skills

Best for: Research Scientist, AI Scientist, Machine Learning Engineer, Data Scientist

Related on AIssential

• Unsupervised Skill Discovery for Agentic Data Analysis — DataCOPE enables unsupervised skill discovery for data-analytic agents using verifier-guided trajectory analysis to improve performance.
• SkillGraph: Graph Foundation Priors for LLM Agent Tool Sequence Recommendation — SkillGraph uses a graph-based prior from agent trajectories to improve LLM tool sequence recommendation beyond semantic similarity.
• SkillRevise: Improving LLM-Authored Agent Skills via Trace-Conditioned Skill Revision — SkillRevise iteratively refines LLM agent skills by diagnosing execution defects, applying principled edits, and empirically validating improvements for optimal performance.
• WebXSkill: Skill Learning for Autonomous Web Agents — WebXSkill bridges the "grounding gap" for web agents by combining executable actions with natural language guidance.
• SkillHone: A Harness for Continual Agent Skill Evolution Through Persistent Decision History — SkillHone enables continuous agent skill evolution by persistently recording and utilizing decision history and evaluation feedback.
• Trace2Skill: Verifier-Guided Skill Evolution for Long-Context EDA Agents — Evolve natural-language agent skills with verifier feedback to solve complex hardware design problems without model fine-tuning.

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by cs.AI updates on arXiv.org.