epicenterhq Radical Options
The local fix might be honoring a bad shape.
The move is simple: step out of the current abstraction before improving it. Sometimes the correct answer is not a cleaner wrapper, a narrower helper, or one more option. Sometimes the correct answer is to redesign the surface from the product sentence down and delete the old constraint that made the code weird.
Related skills: use greenfield-clean-breaks when the radical option changes public contracts, package boundaries, or migration strategy. Use one-sentence-test to name the system before auditing it. Use asymmetric-wins for the refuse-one-shape-to- delete-a-code-family decision. Use post-implementation-review "First-Read Pass" when the problem is mostly first-read clarity.
The Ritual
Do this before editing when the skill triggers.
Current path:
What are we about to do if we stay inside the existing shape?
Friction:
What feels weird, indirect, duplicated, defensive, or hard to explain?
Radical option:
What would the surface look like if designed today with no compatibility burden?
Deletion prize:
What code family disappears if we take the radical option?
User loss:
Who loses what behavior, migration smoothness, or convenience?
Decision:
Take the radical option / keep the current shape
because ...
Write this out when discussing a design or spec. For narrow edits, it can be a short internal pass, but still let the result steer the implementation.
Find The Largest Relevant Unit
Let the shape of the problem determine the frame. Start at the narrow symptom, then move outward until the surrounding unit explains what the change should be:
prose sentence -> paragraph -> section -> document
code expression -> function -> module -> public surface -> workflow
skills rule -> section -> SKILL.md -> skill cluster
Read that unit end to end. Describe how it would look if the new context had always been known, then compare that ideal with the current shape. Choose a local edit only after this comparison shows that it is the coherent result. Do not choose it merely because it minimizes the diff. The pass may confirm a small change or reveal that the surrounding unit should be rewritten.
Start From The Ideal Call Site
Design the consumer experience first. Ignore the existing abstraction for a moment.
const workspace = await openWorkspace({
path,
auth,
});
Then work backward:
What must be true for this call site to be honest?
Which layer should own each invariant?
Which existing abstractions become unnecessary?
Which old behavior is this refusing?
If the ideal call site needs callers to pass unrelated concepts, the boundary is still wrong. If it hides policy the caller must understand, the abstraction is too soft.
Mental Inlining Pass
Mentally inline suspicious helpers, wrappers, options, adapters, and files into their callers.
Ask:
What does this layer add after its name is removed?
Would the caller be clearer with this code inline?
Is this boundary protecting unsafe input or only hiding simple control flow?
Does the abstraction own an invariant, or does it merely pass data through?
Was this created by the old design rather than the current problem?
Keep the abstraction only when it earns one of these jobs:
- owns a real invariant
- isolates unsafe input or interop
- names non-obvious domain behavior
- makes multiple callers simpler in the same way
- gives a package or runtime boundary a stable contract
Otherwise, inline it or redesign above it.
Go Up One Level
When the local fix keeps growing, ask what the code is compensating for.
Local symptom:
We need another boolean to know whether teardown finished.
One level up:
Why can callers observe a half-torn-down workspace?
Radical option:
Make workspace teardown a single owning transition and remove all downstream
"is it gone yet?" checks.
The goal is not a larger diff. The goal is a smaller explanation.
Asymmetric Option Check
Look for one refusal that deletes a whole code family: an old shape, rare mode, fallback, alias, fast path, or provider-specific behavior whose removal collapses adapters, unions, flags, docs branches, tests, UI states, migration code, and defensive checks.
This is the asymmetric wins move. asymmetric-wins owns the candidate list, procedure, and decision template. Default toward refusal when the product sentence survives and the deletion prize is large; keep the behavior when the user loss is load-bearing.
Decision Rules
Choose the radical option when:
- the current abstraction is the main source of complexity
- the higher-level invariant is clear
- refusing the old shape leaves the product sentence intact
- the new system is easier to explain in one sentence
- migration cost is finite and can be done in one clean wave
Keep the current shape when:
- the abstraction owns a real boundary
- the weirdness comes from an external API, file format, or runtime constraint
- the radical option deletes behavior users actually rely on
- the extra migration cost is larger than the explanation win
Output Shape
When reporting back, be direct:
Radical option:
Replace the wrapper stack with one workspace lifecycle owner.
Why:
The wrappers only preserve an old split between auth and workspace state.
What disappears:
Two readiness flags, one adapter, one stale type alias, and three downstream
defensive checks.
Decision:
Take it. The product sentence still holds, and no user-facing behavior is
lost.
Do not make radical options theatrical. The useful version is calm, specific, and willing to delete clever code.