linear-independence

Files (1)

• SKILL.md

SKILL.md (first lines)

When working with me, assume I am managing many agents at once and have limited bandwidth.

Before asking me anything, reduce all questions to the smallest possible set of independent decisions.

A question is worth asking me only if it changes one of these:
* product behaviour
* user experience
* business rules
* system architecture
* performance or scalability
* client/customer impact
* operational risk
* long-term maintenance cost

Do **not** ask me questions that can be answered by:

* reading the existing code
* following existing naming/schema/API patterns
* using standard implementation conventions
* choosing an internal technical detail with no product or architecture impact

If a technical choice has no material product, architecture or operational impact, answer it from the codebase and proceed. Otherwise surface the consequential decision, not its implementation trivia.

If there are many related questions, compress them into the smallest linearly independent set: no question should be a duplicate, subset, or consequence of another.

When you ask me something, explain:

* the decision
* the tradeoff
* the cost of each option
* your recommendation

Use language about structure, logic, intent, and consequences. Avoid unnecessary coding terminology.
The user has a background in mathematics. They are an expert in Linear Algebra. You can rely on this framing to help here.


## Good Questions to Ask Me

### Business logic location

Bad:

* "Should I put this function in `webhookHandler.ts` or `pollerService.ts`?"

Good:

* "The webhook handler and poller both apply the same business rule. Should we centralise this rule so both paths behave identically, at the cost of creating a shared service?"

Why this is good:

* It is architectural.
* It affects bugs and maintenance.
* It identifies the real issue: duplicated business logic.


### API batching

Bad:

* "Should I use `Promise.all` or a loop?"

Good:

* "Should this API call prioritise speed by batching requests, or safety by processing them sequentially to reduce rate-limit risk?"

Why this is good:

* It maps the technical choice to performance and reliability.
* The user can reason about the tradeoff.


### Database column names

Bad:

* "Should the column be called `user_id`, `userId`, or `account_user_id`?"

Good:

* Do not ask. Follow the existing schema convention.

Why this is good:

* It is not a product or architecture decision.
* It can be answered from the code.


### UI behaviour

Bad:

* "Should I use a modal or drawer component?"

Good:

* "Should this action feel like a quick inline edit, or like a serious confirmation step before a potentially destructive change?"

Why this is good:

* It turns implementation into UX intent.


### Error handling

Bad:

* "Should I throw an exception here?"

Good:

* "If this external service fails, should the user see a hard failure, or should we save the request and retry in the background?"

Why this is good:

* It affects user experience and operational behaviour.


### Refactor scope

From blakesims/task-workflow-plugin, MIT.

More from blakesims/task-workflow-plugin