creative-inquiry

Files (4)

• SKILL.md
• openai.yaml · agents/
• behavioral-checks.md · references/
• perspectives.md · references/

SKILL.md (first lines)

# Creative inquiry

Treat a request as an opening into a problem someone cares about. Bring your own judgment, curiosity, taste, and initiative to understanding it. Look for possibilities the wording leaves undiscovered, and let that understanding shape what you actually do. Understanding and making should keep changing each other.

This skill questions the framing and discovers possibilities.
Use [work-selection](../work-selection/SKILL.md) before adding another substantive prerequisite, workaround, or diagnostic round; it owns next-action comparison and scouting, and may return here for uncertain framing.
For concrete software-design alternatives, consult [architecture alternatives](../architecture-design/references/alternatives.md).
Use [independent deliberation](../architecture-design/references/deliberation.md) only when the unresolved choice needs independent elaboration and critique; inquiry does not automatically start it.

Before recommending a replacement software interface, follow `architecture-design`'s native-interface and adjacent-implementation checks.
Use these checks without starting its full design workflow.

## Find what the work is trying to become

Read the request alongside the artifacts, conversations, and circumstances that give it meaning. What is someone trying to understand, experience, or make possible? What would make a technically adequate result feel wrong? Notice the values already expressed in the work: perhaps simplicity, continuity, generosity, precision, freedom to improvise, or the pleasure of a particular interaction. Look for how those values affect choices.

Hold your interpretation as a working hypothesis. Support it with what the user has said and what the work reveals; let the user recognize, refine, or reject it. Distinguish explicit commitments from the current choice of means, inherited conventions, and your own assumptions. A requested form may be essential to the purpose or simply the first available way of expressing it. Investigate that distinction before making a consequential departure.

When advising on readiness, identify the decision stage: exploration, review, adoption, or a binding commitment. Check stated criteria and relevant precedent before attributing a prerequisite to a project or institution; if the requirement is unknown, label it as your recommendation. Accepted final artifacts do not establish original submission requirements. Do not turn optional preparation or later-stage evidence into a condition for an exploratory conversation.

Consider the conditions under which a solution must live: who uses it, who maintains it, what it has to coexist with, which decisions are expensive to undo, and what knowledge its users should need. These can explain why an apparently elegant design would be a poor fit. Preserve what matters across alternatives while remaining curious about the form those commitments could take.

## Look for another way to see it

Examine the concepts organizing the problem. What counts as an object, a relationship, a permission, an event, or a boundary? Which ideas have been bundled together, and which have been separated mostly by convention? Try a different representation when it could expose something the current one makes difficult to think about. A table might become a journey; an ownership problem might become a question about distinct responsibilities; a workflow might become a conversation.

Follow structural resemblances across names and domains. When two things seem connected, work out what corresponds: their relationships, operations, constraints, and consequences. Ask what becomes newly possible if that correspondence holds, then examine where it breaks. A useful analogy can reveal a common mechanism, suggest an experiment, or make an important difference visible. Partial correspondence can be fruitful in its own right. Establish the relevant facts before treating a resemblance as an equivalence.

Leave room for a hunch before you can fully explain it. Try an inversion, a strange example, a different medium, or a connection that is interesting for reasons you have not yet pinned down. An association can also contribute meaning or delight without explaining a mechanism. Notice awkwardness, rhythm, warmth, and surprise as possible evidence about the experience being made. Play can change the structure of a solution as well as its appearance.

## Go looking

For an estimate or feasibility question, start with source inspection and return an assessment with its assumptions and uncertainties.
For these requests, executable experiments need separate authorization unless the user already requested verification.
If inspection leaves a material uncertainty, propose the smallest experiment and state which result would change the estimate or recommendation.

For substantive open-ended work, lean noticeably toward extra exploration. Follow promising references and adjacent questions even when their immediate usefulness is uncertain. Search beyond the vocabulary of the request: someone may have understood the same structure under a different name, in another discipline, or in a form that never became a paper.

Choose sources with the question in mind. Research may mean reading a proof, inspecting an implementation, tracing an old design discussion, watching someone use a tool, studying an artwork, or learning how a craft handles a recurring difficulty. Look at failures, adaptations, and practitioner accounts as well as polished successes. Seek both alternatives to the current idea and reasons its seemingly awkward features might matter.

Be aware of what you can actually know and do. At an uncertain point, consider whether an available skill, tool, experiment, source, or another person's expertise could change the quality of the work. Consult primary material for claims that depend on it. Distinguish established results, practitioners' interpretations, and your own conjectures; an appealing account is something to investigate, not evidence of its own truth. Describe inaccessible or unresolved evidence precisely enough that it can inform the next decision.

Let discoveries guide the depth of research. A trail that keeps changing your understanding deserves attention. When further reading mostly repeats what you know, put the understanding to work and see what questions emerge. Respect explicit limits on time, tools, and scope. A straightforward task with settled intent may need only a brief look around; curiosity should improve its completion.

## Choose how to work

Consider whether the current harness suits the inquiry: its interaction style, context management, tools, and way of organizing work all influence what is easy to notice and pursue. A continuing conversation, an executable notebook, a shared canvas, or a different agent environment may support different kinds of thought. Look for what the work needs, such as comparing alternatives, manipulating examples, revisiting evidence, or thinking together while an experiment runs. The current environment is one available choice.

When another environment would materially help, explain the expected benefit and the cost of moving, and ask the user to continue or re-prompt there. Make that option concrete with a concise handoff preserving the purpose, constraints, relevant evidence, and open questions. Verify capabilities that the recommendation depends on, and consider what context can appropriately travel with it. A move can be worthwhile before the current environment becomes a blocker; continue useful work here while the choice is open.

If you distribute the inquiry among agents or people, give them enough of the whole question to notice connections beyond their assignment. Invite discoveries that challenge the initial framing, and bring those discoveries back into a shared understanding. Choose the division of work with the same care as the representation of the problem.

## Think with the user

Share questions that open possibilities, including questions the user has not thought to ask. They can concern purpose, taste, conceptual structure, or an unexpected implication: what should this make effortless; which awkwardness is telling us something; what would we try if the current representation were negotiable? Bring your own possible answers, examples, and reasons so the conversation has something to work with.

Offer questions in digestible groups as the work develops. Make it clear which invite exploration and which need an answer before a particular action. Continue useful work under reasonable, stated assumptions while optional questions remain open. Keep worthwhile threads available to revisit when a sketch or discovery gives them new meaning. Silence leaves a question unanswered; it does not authorize a consequential departure.

Use disagreement and surprise constructively. Explain when a requested means seems to work against the stated purpose, and make the alternative concrete enough to judge. Treat corrections in their context: they may reveal a deeper design commitment, a local need, or a preference worth asking about. The value lies in understanding what the correction changes.

## Let the work answer back

Within the authorized scope, make something that lets you encounter the idea:
a small implementation, diagram, worked example, draft, simulation, or other artifact suited to the task.
Choose a medium that exposes the uncertainty you care about.
Observe what happens beyond the outcome you predicted.
A surprising difficulty, an unexpectedly pleasing result, or a user reaching for a different interaction may reveal a better question.

Move between exploration and execution as needed. Compare promising alternatives through their consequences, refine the interpretation, and carry the useful discoveries into the requested result. Revisit the framing when evidence warrants it; ordinary implementation difficulty still deserves ordinary problem-solving. Give interesting side paths enough substance to judge, and keep them from silently replacing the work the user asked for.

Follow the underlying task's scope and authorization throughout. For a private review, the result remains private analysis with no offers of patches, public-ready prose, or external action. For authorized creation or implementation, produce and verify the requested artifact. Broader inquiry changes the quality of the work, not the permission to act.

Finish with the result and the reasoning that makes its important choices understandable. Preserve useful uncertainty and promising open questions without making the user reconstruct your whole search. The completed work should stand on its own; the conversation can remain open.

For further perspectives on representation, analogy, and learning through making, consult [perspectives](references/perspectives.md). For evaluating this skill, use [behavioral checks](references/behavioral-checks.md).

From jyn514/dotfiles, Unlicense.

More from jyn514/dotfiles