This is a suite of tools for Research Through Design (RtD) including
a RtD oriented agent and a user persona discovery and elucidation skill.
RtD agent
This agent applies the research through design (RtD) approach to design and engineering problem solving using evidence based approaches. It follows a 5 part framework:
- Frame your own research questions
- Build and test design artifacts (e.g., experiments, prototypes, simulations, processes)
- Use design activity as a method of inquiry
- Produce transferable chemical engineering knowledge
- Ambiguity is intentional. Struggle is expected. Reflection is required.
User Persona related skill
This skill can be used by an agent independently or in a conversation with a user to craft, discovery, and elucidate user personas in a research software engineering or research through design setting.
The skill contains an overview, a when to use section, a section of core princples to guide user persona authorship, and a conversational or independent workflow in 7 phases. Additionally, it has instructions for formatting and styling of outputs in markdown. Finally, it has examples in a reference markdown file.
Core principles include: Evidence based requirements, goal-oriented structure, with behavior as a center-piece, that are archetypical, contextually grounded. The personas should be actionable for subsequent design steps, internally coherent and plausible with explicit acknowledgement of assumptions made in their creation and treated as living artifacts that evolve with the system, technology, and context.
The workflow is conducted in phases:
- Align on purpose
- Share evidence, not opinions
- Affinity clustering and pattern based naming
- Personas as hypotheses
- Stress testing of persona hypotheses
- Surfacing of assumptions
- Reflection and reposition for finalization
The examples in the reference markdown document follow the format: Core Goal: Description Behaviors: Description Constraints: Description Tension: Description Design Implication: Description
This skill has been tested with qwen3:14b, gpt-oss:20b, granite-4.0-h-tiny. It is consistently conversational with qwen and granite. Strong context in AGENTS.md or elsewhere is recommended.