IntentLint: Supporting Intent Scaffolding and Prompt-time Linting in Human-AI Collaborative Data Analysis

Collaborative data analysis is often hindered by the difficulty of maintaining a shared understanding of analytic goals, assumptions, and rationales. As teams increasingly integrate Large Language Models (LLMs) into their workflows, these challenges are amplified. Existing tools often focus on retrospective documentation or artifact production, failing to provide proactive support for coordination. When analytic intent remains implicit, it leads to misaligned strategies, undocumented assumptions, and unwanted agent behaviors that are difficult to debug.

To address these coordination breakdowns, the authors introduce IntentLint, a system that implements a rule-based coordination layer. The system features two primary mechanisms:

• Intent Scaffolding: The system analyzes shared notebooks to infer user goals, assumptions, and rationales, presenting them as structured, editable rules. This allows analysts to externalize their thought processes and make them legible to both human collaborators and AI agents.
• Prompt-time Linting: When a user issues a new prompt, the system checks it against the active, shared rules. If a potential conflict is detected—such as a prompt that contradicts a previously established analytic constraint—the system provides a linting message that highlights the relevant context and suggests revisions before the code is generated.

A study with 16 data analysts demonstrated that IntentLint effectively bridges the gap between intent articulation and action. Participants reported that the system improved their awareness of collaborators' activities and encouraged deeper reflection on their own analytical strategies. By making intent explicit and actionable, IntentLint helps teams identify misalignments early in the workflow, reducing the technical debt and confusion typically associated with multi-human, multi-agent data science projects.