Not every step needs the expensive model
Tier selection is not a preference, it is a function of three facts about the task. Two platform constraints rule out most of the combinations you would guess.
All posts
Bar, rail, drawer — the same destinations, chosen by window width
NavigationBar, NavigationRail and the drawer are three presentations of the same destination list. Window size decides which, and NavigationSuiteScaffold does the switching — with rules about how many destinations belong there at all.
Loading every rule up front is how none of them apply
An always-loaded rulebook competes with itself for attention. Matching the tool target and injecting only what fires turns 634 rules into the twelve that apply.
Two agents, one working tree, one disaster
Parallel agents sharing a working tree corrupt each other silently. Worktrees fix it, and cost 4.3 seconds and 364 MB each — not the 200ms the contract claims.
Reading back is not optional
An agent reporting success has usually verified the call returned, not the artifact. Reading its own output back is the check it skips most often.
The agent whose only job is what is missing
Workers report what they did; nothing reports what nobody did. A pass whose only question is what is missing finds the gaps no individual report can.
Auto-fix needs a floor, not an opinion
Whether an automated repair may apply unattended is arithmetic, not judgement: how certain the fix is against how much it can damage if wrong.
A dialog is state, not a function you call
Dialogs and bottom sheets are composables that exist conditionally, not imperative calls. Modelling them as state fixes the rotation bug, the double-show bug, and the question of where the result goes.
An agent that may only work on the front of the queue
An autonomous drive given a backlog will reach for the interesting item. A head lock forces each item to be dispatched or explicitly dequeued, never skipped.
An evaluator with two outcomes will pick one
A pass-or-fail evaluator forced to rule on insufficient evidence produces a confident guess. A third verdict blocks like a failure while asking for something else.
Stop counting findings and start counting empty rounds
Discovery of unknown size cannot target a count. Convergence on consecutive empty rounds finds the tail that a fixed number of passes always misses.
Say what you truncated or the report is a lie
Top-N, sampling and no-retry all produce reports shaped like full coverage. A bound that is not named in the output is indistinguishable from completeness.