explore

Develop what someone is making into a stronger work, product, or playable experience and hand back something to react to. Use when a reference's effect should move into their own thing, an existing ex

SKILL.md

GitHub

Explore

Carry a goal further than it currently reaches. The person may already know what they are making; Explore's work is to develop it toward the effect they want, into a fuller experience, a stronger work, a playable scene, or a product candidate, and to hand back something they can react to rather than a list of things that could be tried.

Read the goal and what already works

Recover the accepted goal, the qualities it must keep, the references or earlier attempts at hand, and the authority granted. Separate what the person has actually chosen from what an earlier proposal introduced. A reference may be supplied, remembered, or absent; a clear goal neither needs a new brief nor forces a search.

Extract the relationship, not the surface

Look at what a reference does that makes it work, and at the conditions that let it work, rather than copying its surface. The transferable thing is usually a relationship, between effort and reward, tension and release, or choice and consequence, not the skin around it. Read developing-from-references when building from a reference, when the goal is clear but the direction is thin, or when a cross-domain idea has to become concrete.

Then change the conditions on purpose: what if it is longer, slower, shared, riskier, or made for a different player or reader? Moving a condition is how a reference becomes this project instead of a copy.

Inspect the medium that carries the intended effect. If the accepted creative goal depends on sequence or interaction, observe the sequence or actual controls when available and keep text-only or still-only coverage explicit. Do not add motion, controls, randomness, or another mechanism merely to make a probe look substantial.

Let unexpected material open a new focus

Unused material, a half-noticed detail, or something said in passing can be a better opening than the original goal. When a new focus grows from the material, follow it if it still serves the accepted goal, and say why the turn is worth taking. Do not invent a search to look productive, and do not force every observation back into the current frame.

Build the candidate

Turn the direction into something the person can meet: a scene, a rule set, a sample, a sequence, a playable slice. The point is to make a real judgment possible, so enough of the thing must exist for its effect to be felt. Read candidate-and-handoff when producing the candidate, when a prototype seems weak, or when deciding whether the idea or only its execution is at fault.

Do not stop at naming features, offering options, or saying a direction "could work". If the current request authorizes making it and it fits the task, make it and hand back something the person can react to.

Keep what is accepted and keep going

Keep goals the person already accepted; ordinary uncertainty about details is not a reason to reopen them or to ask for new approval, and a goal does not have to be fully settled before Explore can work. When a result does not land, tell a mechanism that fails from an execution too rough to have tested the mechanism, since they call for opposite responses.

Explore carries the creative commission's complete result. When a specialized production method builds the work out, that is the same task continuing within the current authorization, not unfinished work handed back to the user; pass it the candidate and what it still needs to show. Use shared working context when another Skill or worker continues the same line of work.